http://wiki.math.uwaterloo.ca/statwiki/api.php?action=feedcontributions&user=Gchalato&feedformat=atomstatwiki - User contributions [US]2021-08-01T22:39:57ZUser contributionsMediaWiki 1.28.3http://wiki.math.uwaterloo.ca/statwiki/index.php?title=Reinforcement_Learning_of_Theorem_Proving&diff=42356Reinforcement Learning of Theorem Proving2018-12-09T03:05:19Z<p>Gchalato: /* Introduction */</p>
<hr />
<div>== Introduction ==<br />
Automated reasoning over mathematical proofs was a major motivation for the development of computer science. Automated theorem provers(ATPs) can, in principle, be used to attack any formally stated mathematical problem and is a research area that has been present since the early 20th century [1]. As of today, state-of-art ATP systems rely on the fast implementation of complete proof calculi, such as resolution and tableau. However, they are still far weaker than trained mathematicians. Within current ATP systems, many heuristics are essential for their performance. As a result, in recent years machine learning has been used to replace such heuristics and improve the performance of ATPs.<br />
<br />
In this paper, the authors propose a reinforcement learning based ATP, rlCoP. The proposed ATP reasons within first-order logic. The underlying proof calculi are the connection calculi [2], and the reinforcement learning method is Monte Carlo tree search along with policy and value learning. It is shown that reinforcement learning results in a 42.1% performance increase compared to the base prover(without learning).<br />
<br />
== Related Work ==<br />
C. Kalizyk and J. Urban proposed a supervised learning based ATP, FEMaLeCoP, whose underlying proof calculi is the same as this paper in 2015 [3]. Their algorithm learns from existing proofs to choose the next tableau extension step. Since the MaLARea [8] system, number of iterations of a feedback loop between proving and learning have been explored, remarkably improving over human-designed heuristics when reasoning in large theories. However, such systems are known to only learn a high-level selection of relevant facts from a large knowledge base and delegate the internal proof search to standard ATP systems. S. Loos, et al. developed an supervised learning ATP system in 2017 [4], with superposition as their proof calculi. However, they chose deep neural network (CNNs and RNNs) as feature extractor. These systems are treated as black boxes in literature with not much understanding of their performances possible. <br />
<br />
In leanCoP [9], one of the simpler connection tableau systems, the next tableau extension step could be selected using supervised learning. In addition, the first experiments with Monte-Carlo guided proof search [5] have been done for connection tableau systems. The improvement over the baseline measured in that work is much less significant than here. This is closest to the authors' approach but the performance is poorer than this paper.<br />
<br />
On a different note, A. Alemi, et al. proposed a deep sequence model for premise selection in 2016 [6], and they claim to be the first team to involve deep neural networks in ATPs. Although premise selection is not directly linked to automated reasoning, it is still an important component in ATPs, and their paper provides some insights into how to process datasets of formally stated mathematical problems.<br />
<br />
== First Order Logic and Connection Calculi ==<br />
Here we assume basic first-order logic and theorem proving terminology, and we will offer a brief introduction of the bare prover and connection calculi. Let us try to prove the following first-order sentence.<br />
<br />
[[file:fof_sentence.png|frameless|450px|center]]<br />
<br />
This sentence can be transformed into a formula in Skolemized Disjunctive Normal Form (DNF), which is referred to as the "matrix".<br />
<br />
[[file:skolemized_dnf.png|frameless|450px|center]] <br />
[[file:matrix.png|frameless|center]] <br />
<br />
The original first-order sentence is valid if and only if the Skolemized DNF formula is a tautology. The connection calculi attempt to show that the Skolemized DNF formula is a tautology by constructing a tableau. We will start at the special node, root, which is an open leaf. At each step, we select a clause (for example, clause <math display="inline">P \wedge R</math> is selected in the first step), and add the literals as children for an existing open leaf. For every open leaf, examine the path from the root to this leaf. If two literals on this path are unifiable (for example, <math display="inline">Qx'</math> is unifiable with <math display="inline">\neg Qc</math>), this leaf is then closed. An example of a closed tableaux is shown in Figure 1. In standard terminology, it states that a connection is found on this branch.<br />
<br />
[[file:tableaux_example.png|thumb|center|Figure 1. An example of closed tableaux. Adapted from [2]]]<br />
<br />
The paper's goal is to close every leaf, i.e. on every branch, there exists a connection. If such state is reached, the paper has shown that the Skolemized DNF formula is a tautology, thus proving the original first-order sentence. As we can see from the constructed tableaux, the example sentence is indeed valid.<br />
<br />
In formal terms, the rules of connection calculi is shown in Figure 2, and the formal tableaux for the example sentence is shown in Figure 3. Each leaf is denoted as <math display="inline">subgoal, M, path</math> where <math display="inline">subgoal</math> is a list of literals that we need to find connection later, <math display="inline">M</math> stands for the matrix, and <math display="inline">path</math> stands for the path leading to this leaf.<br />
<br />
[[file:formal_calculi.png|thumb|600px|center|Figure 2. Formal connection calculi. Adapted from [2].]]<br />
[[file:formal_tableaux.png|thumb|600px|center|Figure 3. Formal tableaux constructed from the example sentence. Adapted from [2].]]<br />
<br />
To sum up, the bare prover follows a very simple algorithm. given a matrix, a non-negated clause is chosen as the first subgoal. The function ''prove(subgoal, M, path)'' is stated as follows:<br />
* If ''subgoal'' is empty<br />
** return ''TRUE''<br />
* If reduction is possible<br />
** Perform reduction, generating ''new_subgoal'', ''new_path''<br />
** return ''prove(new_subgoal, M, new_path)''<br />
* For all clauses in ''M''<br />
** If a clause can do extension with ''subgoal''<br />
** Perform extension, generating ''new_subgoal1'', ''new_path'', ''new_subgoal2''<br />
** return ''prove(new_subgoal1, M, new_path)'' and ''prove(new_subgoal2, M, path)''<br />
* return ''FALSE''<br />
<br />
It is important to note that the bare prover implemented in this paper is incomplete. Here is a pathological example. Suppose the following matrix (which is trivially a tautology) is feed into the bare prover. Let clause <math display="inline">P(0)</math> be the first subgoal. Clearly choosing <math display="inline">\neg P(0)</math> to extend will complete the proof.<br />
<br />
[[file:pathological.png|frameless|400px|center]] <br />
<br />
However, if we choose <math display="inline">\neg P(x) \lor P(s(x))</math> to do extension, the algorithm will generate an infinite branch <math display="inline">P(0), P(s(0)), P(s(s(0))) ...</math>. It is the task of reinforcement learning to guide the prover in such scenarios towards a successful proof.<br />
<br />
A technique called iterative deepening can be used to avoid such infinite loop, making the bare prover complete. Iterative deepening will force the prover to try all shorter proofs before moving into long ones, it is effective, but also waste valuable computing resource trying to enumerate all short proofs.<br />
<br />
In addition, the provability of first-order sentences is generally undecidable (this result is named the Church-Turing Thesis), which sheds light on the difficulty of automated theorem proving.<br />
<br />
== Mizar Math Library ==<br />
Mizar Math Library (MML) [7, 10] is a library of mathematical theories. The axioms behind the library is the Tarski-Grothendieck set theory, written in first-order logic. The library contains 57,000+ theorems and their proofs, along with many other lemmas, as well as unproven conjectures. Figure 4 shows a Mizar article of the theorem "If <math display="inline"> p </math> is prime, then <math display="inline"> \sqrt p </math> is irrational."<br />
<br />
[[file:mizar_article.png|thumb|center|Figure 4. An article from MML. Adapted from [6].]]<br />
<br />
The training and testing data for this paper is a subset of MML, the Mizar40, which is 32,524 theorems proved by automated theorem provers. Below is an example from the Mizar40 library, it states that with ''d3_xboole_0'' and ''t3_xboole_0'' as premises, we can prove ''t5_xboole_0''.<br />
<br />
[[file:mizar40_0.png|frameless|400px|center]]<br />
[[file:mizar40_1.png|frameless|600px|center]]<br />
[[file:mizar40_2.png|frameless|600px|center]]<br />
[[file:mizar40_3.png|frameless|600px|center]]<br />
<br />
== Monte Carlo Guidance ==<br />
<br />
Monte Carlo tree search (MCTS) is a heuristic search algorithm for some kinds of decision processes. The focus of Monte Carlo tree search is on the analysis of the most promising moves, expanding the search tree based on random sampling of the search space. Then the expansion will then be used to weight the node in the search tree.<br />
<br />
In the reinforcement learning setting, the action is defined as one inference (either reduction or extension). The proof state is defined as the whole tableaux. To implement Monte-Carlo tree search, each proof state <math display="inline"> i </math> needs to maintain three parameters, its prior probability <math display="inline"> p_i </math>, its total reward <math display="inline"> w_i </math>, and number of its visits <math display="inline"> n_i </math>. If no policy learning is used, the prior probabilities are all equal to one. <br />
<br />
A simple heuristic is used to estimate the future reward of leaf states: suppose leaf state <math display="inline"> i </math> has <math display="inline"> G_i </math> open subgoals, the reward is computed as <math display="inline"> 0.95 ^ {G_i} </math>. This will be replaced once value learning is implemented.<br />
<br />
The standard UCT formula is chosen to select the next actions in the playouts<br />
\begin{align}<br />
{\frac{w_i}{n_i}} + 2 \cdot p_i \cdot {\sqrt{\frac{\log N}{n_i}}}<br />
\end{align}<br />
where <math display="inline"> N </math> stands for the total number of visits of the parent node.<br />
<br />
The bare prover is asked to play <math display="inline"> b </math> playouts of length <math display="inline"> d </math> from the empty tableaux, each playout backpropagates the values of proof states it visits. After these <math display="inline"> b </math> playouts a special action (inference) is made, corresponding to an actual move, resulting in a new bigstep tableaux. The next <math display="inline"> b </math> playouts will start from this tableaux, followed by another bigstep, etc.<br />
<br />
== Policy Learning and Guidance ==<br />
<br />
From many runs of MCT, we will know the optimal prior probability of actions (inferences) in particular proof states, we can extract the frequency of each action <math display="inline"> a </math>, and normalize it by dividing with the average action frequency at that state, resulting in a relative proportion <math display="inline"> r_a \in (0, \infty) </math>. We characterize the proof states for policy learning by extracting human-engineered features. Also, we characterize actions by extracting features from the clause chosen and literal chosen as well. Thus we will have a feature vector <math display="inline"> (f_s, f_a) </math>. <br />
<br />
The feature vector <math display="inline"> (f_s, f_a) </math> is regressed against the associated <math display="inline"> r_a </math>.<br />
<br />
During the proof search, the prior probabilities <math display="inline"> p_i </math> of available actions <math display="inline"> a_i </math> in a state <math display="inline"> s </math> is computed as the softmax of their predictions.<br />
<br />
Training examples are only extracted from big step states, making the amount of training data manageable.<br />
<br />
== Value Learning and Guidance ==<br />
<br />
Bigstep states are also used for proof state evaluation. For a proof state <math display="inline"> s </math>, if it corresponds to a successful proof, the value is assigned as <math display="inline"> v_s = 1 </math>. If it corresponds to a failed proof, the value is assigned as <math display="inline"> v_s = 0 </math>. For other scenarios, denote the distance between state <math display="inline"> s </math> and a successful state as <math display="inline"> d_s </math>, then the value is assigned as <math display="inline"> v_s = 0.99^{d_s} </math> <br />
<br />
Proof state feature <math display="inline"> f_s </math> is regressed against the value <math display="inline"> v_s </math>. During the proof search, the reward of leaf states are computed from this prediction.<br />
<br />
== Features and Learners ==<br />
For proof states, features are collected from the whole tableaux (subgoals, matrix, and paths). Each unique symbol is represented by an integer, and the tableaux can be represented as a sequence of integers. Term walk is implemented to combine a sequence of integers into a single integer by multiplying components by a fixed large prime and adding them up. Then the resulting integer is reduced to a smaller feature space by taking modulo by a large prime.<br />
<br />
For actions the feature extraction process is similar, but the term walk is over the chosen literal and the chosen clause.<br />
<br />
In addition to the term walks, they also added several common features: number of goals, total symbol size of all goals, length of active paths, number of current variable instantiations, most common symbols.<br />
<br />
The whole project is implemented in OCaml, and XGBoost is ported into OCaml as the learner.<br />
<br />
== Experimental Results ==<br />
In the paper, the dataset they were using is Mizar40. They divided the mizar40 dataset into training and testing set, with a ratio of 9 to 1. According to the author, the split is a random split. During the experiment, the authors' method was able to prove 32524 statements out of 146700 statements. The authors' main approach is transforming the data from First-order logic form into DNF( disjunctive normal form), <br />
The authors use the M2k dataset to compare the performance of mlCoP, the bare prover and rlCoP using only UCT. There were 577 test problems that the rlCop trained. <br />
*Performance without Learning<br />
Table 3 shows the baseline result. The Performance of the bare prover is significantly lower than mlCoP and rlCoP without policy/value.<br />
[[file:table3.png|550px|center]]<br />
*Reinforcement Learning of Policy Only<br />
In this experiment, the authors evaluated on the dataset rlCoP with UCT using policy learning only. They used the policy training data from previous iterations to train a new predictor after each iteration. Which means only the first iteration ran without policy while all the rest iterations used previous policy training data.<br />
From Table 4, rlCoP is better than mlCoP run with the much higher <math>4 ∗ 10^{6}</math> inference limit after fourth iteration. <br />
[[file:table4.png|550px|center]]<br />
*Reinforcement Learning of Value Only<br />
This experiment was similar to the last one, however, they used only values rather than learned policy. From Table 5, the performance of rlCoP is close to mlCoP but below it after 20 iterations, and it is far below rlCoP using only policy learning.<br />
[[file:table5.png|550px|center]]<br />
*Reinforcement Learning of Policy and Value<br />
From Table 6, the performance of rlCoP is 19.4% more than mlCoP with <math>4 ∗ 10^{6}</math> inferences, 13.6% more than the best iteration of rlCoP with policy only, and 44.3% more than the best iteration of rlCoP with value only after 20 iterations.<br />
[[file:table6.png|550px|center]]<br />
Besides, they also evaluated the effect of the joint reinforcement learning of both policy and value. Replacing final policy and value with the best one from policy-only or value-only both decreased performance.<br />
<br />
*Evaluation on the Whole Miz40 Dataset.<br />
The authors split Mizar40 dataset into 90% training examples and 10% testing examples. 200,000 inferences are allowed for each problem. 10 iterations of policy and value learning are performed (based on MCT). The training and testing results are shown as follows. In the table, ''mlCoP'' represents for the bare prover with iterative deepening (i.e. a complete automated theorem prover with connection calculi), and ''bare prover'' stands for the prover implemented in this paper, without MCT guidance.<br />
<br />
[[file:atp_result0.jpg|frane|550px|center|Figure 5a. Experimental result on Mizar40 dataset]]<br />
[[file:atp_result1.jpg|frame|550px|center|Figure 5b. More experimental result on Mizar40 dataset]]<br />
<br />
As shown by these results, reinforcement learning leads to a significant performance increase for automated theorem proving, the 42.1% performance improvement is unusually high, since the published improvement in this field is typically between 3% and 10%. [1]<br />
<br />
Besides these results, there were also found that some test problems could be solved with rlCoP easily but mlCoP could not.<br />
<br />
[[file:picture3.png|frame|550px|center|Figure 6: The MCTS tree for the WAYBEL 0:28 problem at the moment when the proof is found. For each node we display the predicted probabilityp, the number of visitsnand the average rewardr=w/n. For the (thicker) nodes leading to the proof the corresponding local proof goals arepresented on the right.]]<br />
<br />
== Conclusions ==<br />
In this work, the authors developed an automated theorem prover that uses no domain engineering and instead replies on MCT guided by reinforcement learning. The resulting system is more than 40% stronger than the baseline system. The authors believe that this is a landmark in the field of automated reasoning, demonstrating that building general problem solvers by reinforcement learning is a viable approach. [1]<br />
<br />
The authors pose that some future research could include strong learning algorithms to characterize mathematical data. The development of suitable deep learning architectures will help the algorithm characterize semantic and syntactic features of mathematical objects which will be crucial to create strong assistants for mathematics and hard sciences.<br />
<br />
== Critiques ==<br />
Until now, automated reasoning is relatively new to the field of machine learning, and this paper shows a lot of promise in this research area.<br />
<br />
The feature extraction part of this paper is less than optimal. It is my opinion that with proper neural network architecture, deep learning extracted features will be superior to human-engineered features, which is also shown in [4, 6].<br />
<br />
Also, the policy-value learning iteration is quite inefficient. The learning loop is:<br />
* Loop <br />
** Run MCT with the previous model on an entire dataset<br />
** Collect MCT data<br />
** Train a new model<br />
If we adopt this to an online learning scheme by learning as soon as MCT generates new data, and update the model immediately, there might be some performance increase.<br />
<br />
The experimental design of this paper has some flaws. The authors compare the performance of ''mlCoP'' and ''rlCoP'' by limiting them to the same number of inference steps. However, every inference step of ''rlCoP'' requires additional machine learning prediction, which costs more time. A better way to compare their performance is to set a time limit.<br />
<br />
It would also be interesting to study automated theorem proving in another logic system, like high order logic, because many mathematical concepts can only be expressed in higher-order logic.<br />
<br />
== References ==<br />
[1] C. Kaliszyk, et al. Reinforcement Learning of Theorem Proving. NIPS 2018.<br />
<br />
[2] J. Otten and W. Bibel. leanCoP: Lean Connection-Based Theorem Proving. Journal of Symbolic Computation, vol. 36, pp. 139-161, 2003.<br />
<br />
[3] C. Kaliszyk and J. Urban. FEMaLeCoP: Fairly Efficient Machine Learning Connection Prover. Lecture Notes in Computer Science. vol. 9450. pp. 88-96, 2015.<br />
<br />
[4] S. Loos, et al. Deep Network Guided Proof Search. LPAR-21, 2017.<br />
<br />
[5] M. F¨arber, C. Kaliszyk, and J. Urban. Monte Carlo tableau proof search. In L. de Moura, editor,<br />
26th International Conference on Automated Deduction (CADE), volume 10395 of LNCS,<br />
pages 563–579. Springer, 2017.<br />
<br />
[6] A. Alemi, et al. DeepMath-Deep Sequence Models for Premise Selection. NIPS 2016.<br />
<br />
[7] Mizar Math Library. http://mizar.org/library/<br />
<br />
[8] J. Urban, G. Sutcliffe, P. Pudla ́k, and J. Vyskocˇil. MaLARea SG1 - Machine Learner for Automated Reasoning with Semantic Guidance. In A. Armando, P. Baumgartner, and G. Dowek, editors, IJCAR, volume 5195 of LNCS, pages 441–456. Springer, 2008.<br />
<br />
[9] J. Otten and W. Bibel. leanCoP: lean connection-based theorem proving. J. Symb. Comput., 36(1-2):139–161, 2003.<br />
<br />
[10] A. Grabowski, A. Korniłowicz, and A. Naumowicz. Mizar in a nutshell. J. Formalized Rea-<br />
soning, 3(2):153–245, 2010</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=a_neural_representation_of_sketch_drawings&diff=42355a neural representation of sketch drawings2018-12-09T02:54:06Z<p>Gchalato: /* Introduction */</p>
<hr />
<div><br />
== Introduction ==<br />
In this paper, the authors present a recurrent neural network, sketch-rnn, that can be used to construct stroke-based drawings. Besides new robust training methods, they also outline a framework for conditional and unconditional sketch generation.<br />
<br />
Neural networks have been heavily used as image generation tools. For example, Generative Adversarial Networks, Variational Inference, and Autoregressive models have been used. Most of those models are designed to generate pixels to construct images. People, however, learn to draw using sequences of strokes as opposed to simultaneous generation of pixels. The authors propose a new generative model that creates vector images so that it might generalize abstract concepts in a manner more similar to how humans do. <br />
<br />
The model is trained with hand-drawn sketches as input sequences. The model is able to produce sketches in vector format. In the conditional generation model, they also explore the latent space representation for vector images and discuss a few future applications of this model. The model and dataset are now available as an open source project ([https://magenta.tensorflow.org/sketch_rnn link]).<br />
<br />
=== Terminology ===<br />
Pixel images, also referred to as raster or bitmap images are files that encode image data as a set of pixels. These are the most common image type, with extensions such as .png, .jpg, .bmp. <br />
<br />
Vector images are files that encode image data as paths between points. SVG and EPS file types are used to store vector images. <br />
<br />
For a visual comparison of raster and vector images, see this [https://www.youtube.com/watch?v=-Fs2t6P5AjY video]. As mentioned, vector images are generally simpler and more abstract, whereas raster images generally are used to store detailed images. <br />
<br />
For this paper, the important distinction between the two is that the encoding of images in the model will be inherently more abstract because of the vector representation. The intuition is that generating abstract representations is more effective using a vector representation.<br />
<br />
== Related Work ==<br />
There are some works in the history that used a similar approach to generate images such as Portrait Drawing by Paul the Robot [26, 28] and some reinforcement learning approaches[28], Reinforcement Learning to discover a set of paint brush strokes that can best represent a given input photograph. They work more like a mimic of digitized photographs. There are also some Neural networks based approaches, but those are mostly dealing with pixel images. Little work is done on vector images generation. There are models that use Hidden Markov Models [25] or Mixture Density Networks [2] to generate human sketches, continuous data points (modelling Chinese characters as a sequence of pen stroke actions) or vectorized Kanji characters [9,29].<br />
<br />
Neural Network-based approaches are able to generate latent space representation of vector images, which follows a Gaussian distribution. The generated output of these networks is trained to match the Gaussian distribution by minimizing a given loss function. Using this idea, previous works attempted to generate a sequence-to-Sequence model with Variational Autoencoder to model sentences into latent space and using probabilistic program induction to model Omniglot dataset. Variational Autoencoders differ from regular encoders in that there is an intermediary “sampling step” between the encoder and decoder. Simply connecting the two would NOT guarantee that encoded parameters can be viewed as parameters of a normal distribution representing a latent space. In VAEs, the output of the encoder is physically put into an intermediary step that uses it as normal parameters, and provides a sample. In this way, the encoding is penalized as if it were the parameters of some Normal Distribution.<br />
<br />
One of the limiting factors that the authors mention in the field of generative vector drawings is the lack of availability of publicly available datasets. Previous datasets such as the Sketch data with 20k vector sketches was explored for feature extraction techniques. The Sketchy dataset consisting of 70k vector sketches along with pixel images was used for large-scale exploration of human sketches. The ShadowDraw system that used 30k raster images along with extracted vectorized features is an interactive system<br />
that predicts what a finished drawing looks like based on a set of incomplete brush strokes from the<br />
user while the sketch is being drawn. In all the cases, the datasets are comparatively small. The dataset proposed in this work uses a much larger dataset and has been made publicly available, and is one of the major contributions of this paper.<br />
<br />
== Major Contributions ==<br />
This paper makes the following major contributions: Authors outline a framework for both unconditional and<br />
conditional generation of vector images composed of a sequence of lines. The recurrent neural<br />
network-based generative model is capable of producing sketches of common objects in a vector<br />
format. The paper develops a training procedure unique to vector images to make the training more robust. The paper also made available<br />
a large dataset of hand drawn vector images to encourage further development of generative modelling<br />
for vector images, and also release an implementation of our model as an open source project<br />
<br />
== Methodology ==<br />
=== Dataset ===<br />
QuickDraw is a dataset with 50 million vector drawings collected by an online game [https://quickdraw.withgoogle.com/# Quick Draw!], where the players are required to draw objects belonging to a particular object class in less than 20 seconds. It contains hundreds of classes, each class has 70k training samples, 2.5k validation samples and 2.5k test samples.<br />
<br />
The data format of each sample is a representation of a pen stroke action event. The Origin is the initial coordinate of the drawing. The sketches are points in a list. Each point consists of 5 elements <math> (\Delta x, \Delta y, p_{1}, p_{2}, p_{3})</math> where x and y are the offset distance in x and y directions from the previous point. The parameters <math>p_{1}, p_{2}, p_{3}</math> represent three possible states in binary one-hot representation where <math>p_{1}</math> indicates the pen is touching the paper, <math>p_{2}</math> indicates the pen will be lifted from here, and <math>p_{3}</math> represents the drawing has ended.<br />
<br />
=== Sketch-RNN ===<br />
[[File:sketchfig2.png|700px|center]]<br />
<br />
The model is a Sequence-to-Sequence Variational Autoencoder(VAE). <br />
<br />
==== Encoder ====<br />
The encoder is a bidirectional RNN. The input is a sketch sequence denoted by <math>S =\{S_0, S_1, ... S_{N_{s}}\}</math> and a reversed sketch sequence denoted by <math>S_{reverse} = \{S_{N_{s}},S_{N_{s}-1}, ... S_0\}</math>. The final hidden layer representations of the two encoded sequences <math>(h_{ \rightarrow}, h_{ \leftarrow})</math> are concatenated to form a latent vector, <math>h</math>, of size <math>N_{z}</math>,<br />
<br />
\begin{split}<br />
&h_{ \rightarrow} = encode_{ \rightarrow }(S), \\<br />
&h_{ \leftarrow} = encode_{ \leftarrow }(S_{reverse}), \\<br />
&h = [h_{\rightarrow}; h_{\leftarrow}].<br />
\end{split}<br />
<br />
Then the authors project <math>h</math> into two vectors <math>\mu</math> and <math>\hat{\sigma}</math> of size <math>N_{z}</math>. The projection is performed using a fully connected layer. These two vectors are the parameters of the latent space Gaussian distribution that will estimate the distribution of the input data. Because standard deviations cannot be negative, an exponential function is used to convert it to all positive values. Next, a random variable with mean <math>\mu</math> and standard deviation <math>\sigma</math> is constructed by scaling a normalized IID Gaussian, <math>\mathcal{N}(0,I)</math>, <br />
<br />
\begin{split}<br />
& \mu = W_\mu h + b_\mu, \\<br />
& \hat \sigma = W_\sigma h + b_\sigma, \\<br />
& \sigma = exp( \frac{\hat \sigma}{2}), \\<br />
& z = \mu + \sigma \odot \mathcal{N}(0,I). <br />
\end{split}<br />
<br />
<br />
Note that <math>z</math> is not deterministic but a random vector that can be conditioned on an input sketch sequence.<br />
<br />
==== Decoder ====<br />
The decoder is an autoregressive RNN. The initial hidden and cell states are generated using <math>[h_0;c_0] = \tanh(W_z z + b_z)</math>. Here, <math>c_0</math> is utilized if applicable (eg. if an LSTM decoder is used). <math>S_0</math> is defined as <math>(0,0,1,0,0)</math> (the pen is touching the paper at location 0, 0). <br />
<br />
For each step <math>i</math> in the decoder, the input <math>x_i</math> is the concatenation of the previous point <math>S_{i-1}</math> and the latent vector <math>z</math>. The outputs of the RNN decoder <math>y_i</math> are parameters for a probability distribution that will generate the next point <math>S_i</math>. <br />
<br />
The authors model <math>(\Delta x,\Delta y)</math> as a Gaussian mixture model (GMM) with <math>M</math> normal distributions and model the ground truth data <math>(p_1, p_2, p_3)</math> as a categorical distribution <math>(q_1, q_2, q_3)</math> where <math>q_1, q_2\ \text{and}\ q_3</math> sum up to 1,<br />
<br />
\begin{align*}<br />
p(\Delta x, \Delta y) = \sum_{j=1}^{M} \Pi_j \mathcal{N}(\Delta x,\Delta y | \mu_{x,j}, \mu_{y,j}, \sigma_{x,j},\sigma_{y,j}, \rho _{xy,j}), where \sum_{j=1}^{M}\Pi_j = 1<br />
\end{align*}<br />
<br />
Where <math>\mathcal{N}(\Delta x,\Delta y | \mu_{x,j}, \mu_{y,j}, \sigma_{x,j},\sigma_{y,j}, \rho _{xy,j})</math> is a bi-variate Normal Distribution, with parameters means <math>\mu_x, \mu_y</math>, standard deviations <math>\sigma_x, \sigma_y</math> and correlation parameter <math>\rho_{xy}</math>. There are <math>M</math> such distributions. <math>\Pi</math> is a categorical distribution vector of length <math>M</math>. Collectively these form the mixture weights of the Gaussian Mixture model.<br />
<br />
The output vector <math>y_i</math> is generated using a fully-connected forward propagation in the hidden state of the RNN.<br />
<br />
\begin{split}<br />
&x_i = [S_{i-1}; z], \\<br />
&[h_i; c_i] = forward(x_i,[h_{i-1}; c_{i-1}]), \\<br />
&y_i = W_y h_i + b_y, \\<br />
&y_i \in \mathbb{R}^{6M+3}. \\<br />
\end{split}<br />
<br />
The output consists the probability distribution of the next data point.<br />
<br />
\begin{align*}<br />
[(\hat\Pi_1\ \mu_x\ \mu_y\ \hat\sigma_x\ \hat\sigma_y\ \hat\rho_{xy})_1\ (\hat\Pi_1\ \mu_x\ \mu_y\ \hat\sigma_x\ \hat\sigma_y\ \hat\rho_{xy})_2\ ...\ (\hat\Pi_1\ \mu_x\ \mu_y\ \hat\sigma_x\ \hat\sigma_y\ \hat\rho_{xy})_M\ (\hat{q_1}\ \hat{q_2}\ \hat{q_3})] = y_i<br />
\end{align*}<br />
<br />
<math>\exp</math> and <math>\tanh</math> operations are applied to ensure that the standard deviations are non-negative and the correlation value is between -1 and 1.<br />
<br />
\begin{align*}<br />
\sigma_x = \exp (\hat \sigma_x),\ <br />
\sigma_y = \exp (\hat \sigma_y),\ <br />
\rho_{xy} = \tanh(\hat \rho_{xy}). <br />
\end{align*}<br />
<br />
Categorical distribution probabilities for <math>(p_1, p_2, p_3)</math> using <math>(q_1, q_2, q_3)</math> can be obtained as :<br />
<br />
\begin{align*}<br />
q_k = \frac{\exp{(\hat q_k)}}{ \sum\nolimits_{j = 1}^{3} \exp {(\hat q_j)}},<br />
k \in \left\{1,2,3\right\}, <br />
\Pi _k = \frac{\exp{(\hat \Pi_k)}}{ \sum\nolimits_{j = 1}^{M} \exp {(\hat \Pi_j)}},<br />
k \in \left\{1,...,M\right\}.<br />
\end{align*}<br />
<br />
It is hard for the model to decide when to stop drawing because the probabilities of the three events <math>(p_1, p_2, p_3)</math> are very unbalanced. Researchers in the past have used different weights for each pen event probability, but the authors found this approach lacking elegance and inadequate. They define a hyperparameter representing the max length of the longest sketch in the training set denoted by <math>N_{max}</math>, and set the <math>S_i</math> to be <math>(0, 0, 0, 0, 1)</math> for <math>i > N_s</math>.<br />
<br />
The outcome sample <math>S_i^{'}</math> can be generated in each time step during sample process and fed as input for the next time step. The process will stop when <math>p_3 = 1</math> or <math>i = N_{max}</math>. The output is not deterministic but conditioned random sequences. The level of randomness can be controlled using a temperature parameter <math>\tau</math>.<br />
<br />
\begin{align*}<br />
\hat q_k \rightarrow \frac{\hat q_k}{\tau}, <br />
\hat \Pi_k \rightarrow \frac{\hat \Pi_k}{\tau}, <br />
\sigma_x^2 \rightarrow \sigma_x^2\tau, <br />
\sigma_y^2 \rightarrow \sigma_y^2\tau. <br />
\end{align*}<br />
<br />
The <math>\tau</math> ranges from 0 to 1. When <math>\tau = 0</math> the output will be deterministic as the sample will consist of the points on the peak of the probability density function.<br />
<br />
=== Unconditional Generation ===<br />
There is a special case that only the decoder RNN module is trained. The decoder RNN could work as a standalone autoregressive model without latent variables. In this case, initial states are 0, the input <math>x_i</math> is only <math>S_{i-1}</math> or <math>S_{i-1}^{'}</math>. In the Figure 3, generating sketches unconditionally from the temperature parameter <math>\tau = 0.2</math> at the top in blue, to <math>\tau = 0.9</math> at the bottom in red.<br />
<br />
[[File:sketchfig3.png|700px|center]]<br />
<br />
=== Training ===<br />
The training process is the same as a Variational Autoencoder. The loss function is the sum of Reconstruction Loss <math>L_R</math> and the Kullback-Leibler Divergence Loss <math>L_{KL}</math>. The reconstruction loss <math>L_R</math> can be obtained with generated parameters of pdf and training data <math>S</math>. It is the sum of the <math>L_s</math> and <math>L_p</math>, which are the log loss of the offset <math>(\Delta x, \Delta y)</math> and the pen state <math>(p_1, p_2, p_3)</math>.<br />
<br />
\begin{align*}<br />
L_s = - \frac{1 }{N_{max}} \sum_{i = 1}^{N_s} \log(\sum_{i = 1}^{M} \Pi_{j,i} \mathcal{N}(\Delta x,\Delta y | \mu_{x,j,i}, \mu_{y,j,i}, \sigma_{x,j,i},\sigma_{y,j,i}, \rho _{xy,j,i})), <br />
\end{align*}<br />
\begin{align*}<br />
L_p = - \frac{1 }{N_{max}} \sum_{i = 1}^{N_{max}} \sum_{k = 1}^{3} p_{k,i} \log (q_{k,i}), <br />
L_R = L_s + L_p.<br />
\end{align*}<br />
<br />
<br />
Both terms are normalized by <math>N_{max}</math>.<br />
<br />
<math>L_{KL}</math> measures the difference between the distribution of the latent vector <math>z</math> and an i.i.d. Gaussian vector with zero mean and unit variance.<br />
<br />
\begin{align*}<br />
L_{KL} = - \frac{1}{2 N_z} (1+\hat \sigma - \mu^2 - \exp(\hat \sigma))<br />
\end{align*}<br />
<br />
The overall loss is weighted as:<br />
<br />
\begin{align*}<br />
Loss = L_R + w_{KL} L_{KL}<br />
\end{align*}<br />
<br />
When <math>w_{KL} = 0</math>, the model becomes a standalone unconditional generator. Specially, there will be no <math>L_{KL} </math> term as we only optimize for <math>L_{R} </math>. By removing the <math>L_{KL} </math> term the model approaches a pure autoencoder, meaning it sacrifices the ability to enforce a prior over the latent space and gains better reconstruction loss metrics.<br />
<br />
While the aforementioned loss function could be used, it was found that annealing the KL term (as shown below) in the loss function produces better results.<br />
<br />
<center><math><br />
\eta_{step} = 1 - (1 - \eta_{min})R^{step}<br />
</math></center><br />
<br />
<center><math><br />
Loss_{train} = L_R + w_{KL} \eta_{step} max(L_{KL}, KL_{min})<br />
</math></center><br />
<br />
As shown in Figure 4, the <math>L_{R} </math> metric for the standalone decoder model is actually an upper bound for different models using a latent vector. The reason is the unconditional model does not access to the entire sketch it needs to generate.<br />
<br />
[[File:s.png|600px|thumb|center|Figure 4. Tradeoff between <math>L_{R} </math> and <math>L_{KL} </math>, for two models trained on single class datasets (left).<br />
Validation Loss Graph for models trained on the Yoga dataset using various <math>w_{KL} </math>. (right)]]<br />
<br />
== Experiments ==<br />
The authors experiment with the sketch-rnn model using different settings and recorded both losses. They used a Long Short-Term Memory(LSTM) model as an encoder and a HyperLSTM as a decoder. HyperLSTM is a type of RNN cell that excels at sequence generation tasks. The ability for HyperLSTM to spontaneously augment its own weights enables it to adapt to many different regimes<br />
in a large diverse dataset. They also conduct multi-class datasets. The result is as follows.<br />
<br />
[[File:sketchtable1.png|700px|center]]<br />
<br />
We could see the trade-off between <math>L_R</math> and <math>L_{KL}</math> in this table clearly. Furthermore, <math>L_R</math> decreases as <math>w_{KL} </math> is halfed. <br />
<br />
=== Conditional Reconstruction ===<br />
The authors assess the reconstructed sketch with a given sketch with different <math>\tau</math> values. We could see that with high <math>\tau</math> value on the right, the reconstructed sketches are more random.<br />
<br />
[[File:sketchfig5.png|700px|center]]<br />
<br />
They also experiment on inputting a sketch from a different class. The output will still keep some features from the class that the model is trained on.<br />
<br />
=== Latent Space Interpolation ===<br />
The authors visualize the reconstruction sketches while interpolating between latent vectors using different <math>w_{KL}</math> values. With high <math>w_{KL}</math> values, the generated images are more coherently interpolated.<br />
<br />
[[File:sketchfig6.png|700px|center]]<br />
<br />
=== Sketch Drawing Analogies ===<br />
Since the latent vector <math>z</math> encode conceptual features of a sketch, those features can also be used to augment other sketches that do not have these features. This is possible when models are trained with low <math>L_{KL}</math> values. The authors are able to perform vector arithmetic on latent vectors from different sketches and explore how the model generates sketches base on these latent spaces.<br />
<br />
=== Predicting Different Endings of Incomplete Sketches === <br />
This model is able to predict an incomplete sketch by encoding the sketch into hidden state <math>h</math> using the decoder and then using <math>h</math> as an initial hidden state to generate the remaining sketch. The authors train on individual classes by using decoder-only models and set <math>τ = 0.8</math> to complete samples. Figure 7 shows the results.<br />
<br />
[[File:sketchfig7.png|700px|center]]<br />
<br />
== Limitations ==<br />
<br />
Although sketch-rnn can model a large variety of sketch drawings, there are several limitations in the current approach. For most single-class datasets, sketch-rnn is capable of modelling around 300 data points. The model becomes increasingly difficult to train beyond this length. For the author's dataset, the Ramer-Douglas-Peucker algorithm is used to simplify the strokes of sketch data to less than 200 data points.<br />
<br />
For more complicated classes of images, such as mermaids or lobsters, the reconstruction loss metrics are not as good compared to simpler classes such as ants, faces or firetrucks. The models trained on these more challenging image classes tend to draw smoother, more circular line segments that do not resemble individual sketches, but rather resemble an averaging of many sketches in the training set. This smoothness may be analogous to the blurriness effect produced by a Variational Autoencoder that is trained on pixel images. Depending on the use case of the model, smooth circular lines can be viewed as aesthetically pleasing and a desirable property.<br />
<br />
While both conditional and unconditional models are capable of training on datasets of several classes, sketch-rnn is ineffective at modelling a large number of classes simultaneously. The samples generated will be incoherent, with different classes are shown in the same sketch.<br />
<br />
== Applications and Future Work ==<br />
The authors believe this model can assist artists by suggesting how to finish a sketch, helping them to find interesting intersections between different drawings or objects, or generating a lot of similar but different designs. In the simplest use, pattern designers can apply sketch-rnn to generate a large number of similar, but unique designs for textile or wallpaper prints. The creative designers can also come up with abstract designs which enables them to resonate more with their target audience<br />
<br />
This model may also find its place on teaching students how to draw. Even with the simple sketches in QuickDraw, the authors of this work have become much more proficient at drawing animals, insects, and various sea creatures after conducting these experiments. <br />
When the model is trained with a high <math>w_{KL}</math> and sampled with a low <math>\tau</math>, it may help to turn a poor sketch into a more aesthetical one. Latent vector augmentation could also help to create a better drawing by inputting user-rating data during training processes.<br />
<br />
The authors conclude by providing the following future directions to this work:<br />
# Investigate using user-rating data to augmenting the latent vector in the direction that maximizes the aesthetics of the drawing.<br />
# Look into combining variations of sequence-generation models with unsupervised, cross-domain pixel image generation models.<br />
<br />
It's exciting that they manage to combine this model with other unsupervised, cross-domain pixel image generation models to create photorealistic images from sketches.<br />
<br />
The authors have also mentioned the opposite direction of converting a photograph of an object into an unrealistic, but similar looking<br />
sketch of the object composed of a minimal number of lines to be a more interesting problem.<br />
<br />
Moreover, it would be interesting to see how varying loss will be represented as a drawing. Some exotic form of loss function may change the way that the network behaves, which can lead to various applications.<br />
<br />
== Conclusion ==<br />
The paper presents a methodology to model sketch drawings using recurrent neural networks. The sketch-rnn model that can encode and decode sketches, generate and complete unfinished sketches is introduced in this paper. In addition, Authors demonstrated how to both interpolate between latent spaces from a different class, and use it to augment sketches or generate similar looking sketches. Furthermore, the importance of enforcing a prior distribution on latent vector while interpolating coherent sketch generations is shown. Finally, a large sketch drawings dataset for future research work is created.<br />
<br />
== Critique ==<br />
This paper presents both a novel large dataset of sketches and a new RNN architecture to generate new sketches. It is very exciting to read but there are still some aspect to improve.<br />
<br />
* The performance of the decoder model can hardly be evaluated. The authors present the performance of the decoder by showing the generated sketches, it is clear and straightforward, however, not very efficient. It would be great if the authors could present a way, or a metric to evaluate how well the sketches are generated rather than printing them out and evaluate with human judgment. The authors didn't present an evaluation of the algorithms either. They provided <math>L_R</math> and <math>L_{KL}</math> for reference, however, a lower loss doesn't represent a better performance. Training loss alone likely does not capture the quality of a sketch.<br />
<br />
* The authors have not mentioned details on training details such as learning rate, training time, parameter size, and so on. <br />
<br />
* Algorithm lacks comparison to the prior state of the art on standard metrics, which made the novelty unclear. Using strokes as inputs is a novel and innovative move, however, the paper does not provide a baseline or any comparison with other methods or algorithms. Some other researches were mentioned in the paper, using similar and smaller datasets. It would be great if the authors could use some basic or existing methods a baseline and compare with the new algorithm.<br />
<br />
* Besides the comparison with other algorithms, it would also be great if the authors could remove or replace some component of the algorithm in the model to show if one part is necessary, or what made them decide to include a specific component in the algorithm.<br />
<br />
* The authors did not present better complexity and deeper mathematical analysis on the algorithms in the paper. It also does not include comparison using some more standard metrics compare to previous results. Therefore, it lacks some algorithmic contribution. It would be better to include some more formal analysis on the algorithmic side. <br />
<br />
* The authors proposed a few future applications for the model, however, the current output seems somehow not very close to their descriptions. But I do believe that this is a very good beginning, with the release of the sketch dataset, it must attract more scholars to research and improve with it!<br />
<br />
* As they said their model can become increasingly difficult to train on with increased size.<br />
<br />
== References == <br />
# Jimmy L. Ba, Jamie R. Kiros, and Geoffrey E. Hinton. Layer normalization. NIPS, 2016.<br />
# Christopher M. Bishop. Mixture density networks. Technical Report, 1994. URL http://publications.aston.ac.uk/373/.<br />
# Samuel R. Bowman, Luke Vilnis, Oriol Vinyals, Andrew M. Dai, Rafal Józefowicz, and Samy Bengio. Generating Sentences from a Continuous Space. CoRR, abs/1511.06349, 2015. URL http://arxiv.org/abs/1511.06349.<br />
# H. Dong, P. Neekhara, C. Wu, and Y. Guo. Unsupervised Image-to-Image Translation with Generative Adversarial Networks. ArXiv e-prints, January 2017.<br />
# David H. Douglas and Thomas K. Peucker. Algorithms for the reduction of the number of points required to represent a digitized line or its caricature. Cartographica: The International Journal for Geographic Information and Geovisualization, 10(2):112–122, October 1973. doi: 10.3138/fm57-6770-u75u-7727. URL http://dx.doi.org/10.3138/fm57-6770-u75u-7727.<br />
# Mathias Eitz, James Hays, and Marc Alexa. How Do Humans Sketch Objects? ACM Trans. Graph.(Proc. SIGGRAPH), 31(4):44:1–44:10, 2012.<br />
# I. Goodfellow. NIPS 2016 Tutorial: Generative Adversarial Networks. ArXiv e-prints, December 2016.<br />
# Alex Graves. Generating sequences with recurrent neural networks. arXiv:1308.0850, 2013.<br />
# David Ha. Recurrent Net Dreams Up Fake Chinese Characters in Vector Format with TensorFlow, 2015.<br />
# David Ha, Andrew M. Dai, and Quoc V. Le. HyperNetworks. In ICLR, 2017.<br />
# Sepp Hochreiter and Juergen Schmidhuber. Long short-term memory. Neural Computation, 1997.<br />
# P. Isola, J.-Y. Zhu, T. Zhou, and A. A. Efros. Image-to-Image Translation with Conditional Adversarial Networks. ArXiv e-prints, November 2016.<br />
# Jonas Jongejan, Henry Rowley, Takashi Kawashima, Jongmin Kim, and Nick Fox-Gieg. The Quick, Draw! - A.I. Experiment. https://quickdraw.withgoogle.com/, 2016. URL https: //quickdraw.withgoogle.com/.<br />
# C. Kaae Sønderby, T. Raiko, L. Maaløe, S. Kaae Sønderby, and O. Winther. Ladder Variational Autoencoders. ArXiv e-prints, February 2016.<br />
# T. Kim, M. Cha, H. Kim, J. Lee, and J. Kim. Learning to Discover cross-domain Relations with Generative Adversarial Networks. ArXiv e-prints, March 2017.<br />
# D. P Kingma and M. Welling. Auto-Encoding Variational Bayes. ArXiv e-prints, December 2013.<br />
# Diederik Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In ICLR, 2015.<br />
# Diederik P. Kingma, Tim Salimans, and Max Welling. Improving variational inference with inverse autoregressive flow. CoRR, abs/1606.04934, 2016. URL http://arxiv.org/abs/1606.04934.<br />
# Brenden M. Lake, Ruslan Salakhutdinov, and Joshua B. Tenenbaum. Human level concept learning through probabilistic program induction. Science, 350(6266):1332–1338, December 2015. ISSN 1095-9203. doi: 10.1126/science.aab3050. URL http://dx.doi.org/10.1126/science.aab3050.<br />
# Yong Jae Lee, C. Lawrence Zitnick, and Michael F. Cohen. Shadowdraw: Real-time user guidance for freehand drawing. In ACM SIGGRAPH 2011 Papers, SIGGRAPH ’11, pp. 27:1–27:10, New York, NY, USA, 2011. ACM. ISBN 978-1-4503-0943-1. doi: 10.1145/1964921.1964922. URL http://doi.acm.org/10.1145/1964921.1964922.<br />
# M.-Y. Liu, T. Breuel, and J. Kautz. Unsupervised Image-to-Image Translation Networks. ArXiv e-prints, March 2017.<br />
# S. Reed, A. van den Oord, N. Kalchbrenner, S. Gómez Colmenarejo, Z. Wang, D. Belov, and N. de Freitas. Parallel Multiscale Autoregressive Density Estimation. ArXiv e-prints, March 2017.<br />
# Patsorn Sangkloy, Nathan Burnell, Cusuh Ham, and James Hays. The Sketchy Database: Learning to Retrieve Badly Drawn Bunnies. ACM Trans. Graph., 35(4):119:1–119:12, July 2016. ISSN 0730-0301. doi: 10.1145/2897824.2925954. URL http://doi.acm.org/10.1145/2897824.2925954.<br />
# Mike Schuster, Kuldip K. Paliwal, and A. General. Bidirectional recurrent neural networks. IEEE Transactions on Signal Processing, 1997.<br />
# Saul Simhon and Gregory Dudek. Sketch interpretation and refinement using statistical models. In Proceedings of the Fifteenth Eurographics Conference on Rendering Techniques, EGSR’04, pp. 23–32, Aire-la-Ville, Switzerland, Switzerland, 2004. Eurographics Association. ISBN 3-905673-12-6. doi: 10.2312/EGWR/EGSR04/023-032. URL http://dx.doi.org/10.2312/EGWR/EGSR04/023-032.<br />
# Patrick Tresset and Frederic Fol Leymarie. Portrait drawing by paul the robot. Comput. Graph.,37(5):348–363, August 2013. ISSN 0097-8493. doi: 10.1016/j.cag.2013.01.012. URL http://dx.doi.org/10.1016/j.cag.2013.01.012.<br />
# T. White. Sampling Generative Networks. [https://arxiv.org/abs/1609.04468 ArXiv e-prints], September 2016.<br />
#Ning Xie, Hirotaka Hachiya, and Masashi Sugiyama. Artist agent: A reinforcement learning approach to automatic stroke generation in oriental ink painting. In ICML. icml.cc / Omnipress, 2012. URL http://dblp.uni-trier.de/db/conf/icml/icml2012.html#XieHS12.<br />
# Xu-Yao Zhang, Fei Yin, Yan-Ming Zhang, Cheng-Lin Liu, and Yoshua Bengio. Drawing and Recognizing Chinese Characters with Recurrent Neural Network. CoRR, abs/1606.06539, 2016. URL http://arxiv.org/abs/1606.06539.</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=policy_optimization_with_demonstrations&diff=42354policy optimization with demonstrations2018-12-09T02:48:03Z<p>Gchalato: </p>
<hr />
<div>= Introduction =<br />
<br />
Reinforcement learning (RL) method has led to state-of-the-art results in a variety of applications. However, problems that involve learning from novel policies (to improve long-term performance) still pose challenges - especially in environments where reward signals are sparse and rare. There are currently two ways to solve such exploration problems in RL: <br />
<br />
1) Guide the agent to explore states that have never been seen. <br />
<br />
2) Guide the agent to imitate a demonstration trajectory sampled from an expert policy to learn. <br />
<br />
When guiding the agent to imitate the expert behavior for learning, there are also two methods: putting the demonstration directly into the replay memory [1] [2] [3] or using the demonstration trajectory to pre-train the policy in a supervised manner [4]. However, neither of these methods takes full advantage of the demonstration data. They instead treat the demonstration data identically to self-generated data, requiring a tremendous number of difficult to collect examples to learn effectively. To address this problem, a novel policy optimization method from demonstration (POfD) is proposed, which takes full advantage of the demonstration and there is no need to ensure that the expert policy is the optimal policy. To summarize, the authors bring forth this idea through the following techniques:<br />
<br />
1) A demonstration guided exploration term measuring the divergence between current and the expert policy is added to the policy optimization objective, increasing the similarity to expert-like exploration.<br />
<br />
2) They say that for better learning from demonstrations and getting an optimization friendly lower bound, the proposed objective could be defined on an occupancy measure as in [14].<br />
<br />
3) Finally, they show that the optimization can move towards optimizing the derived lower bound and the generative adversarial training.<br />
<br />
The authors also evaluate the performance of POfD on Mujoco [5] in sparse-reward environments. The experiments results show that the performance of POfD is greatly improved compared with some strong baselines and even to the policy gradient method in dense-reward environments.<br />
<br />
==Intuition==<br />
The agent should imitate the demonstrated behavior when rewards are sparse and then explore new states on its own after acquiring sufficient skills, which is a dynamic intrinsic reward mechanism that can be reshaped in terms of the native rewards in RL. At present, the state of the art exploration in Reinforcement learning is simply epsilon greedy which just makes random moves for a small percentage of times to explore unexplored moves. This is very naive and is one of the main reasons for the high sample complexity in RL. On the other hand, if there is an expert demonstrator who can guide exploration, the agent can make more guided and accurate exploratory moves.<br />
<br />
=Related Work =<br />
There are some related works in overcoming exploration difficulties by learning from demonstration [6] and imitation learning in RL.<br />
<br />
For learning from demonstration (LfD),<br />
# Most LfD methods adopt value-based RL algorithms, such as DQfD (Deep Q-learning from Demonstrations) [2] which are applied into the discrete action spaces and DDPGfD (Deep Deterministic Policy Gradient from Demonstrations) [3] which extends this idea to the continuous spaces. But both of them under-utilize the demonstration data.<br />
# There are some methods based on policy iteration [7] [8], which shapes the value function by using demonstration data. But they get the bad performance when demonstration data is imperfect.<br />
# A hybrid framework [9] that learns the policy in which the probability of taking demonstrated actions is maximized is proposed, which considers fewer demonstration data.<br />
# A reward reshaping mechanism [10] that encourages taking actions close to the demonstrated ones is proposed. It is similar to the method in this paper, but there exist some differences as it is defined as a potential function based on multi-variate Gaussian to model the distribution of state-actions.<br />
All of the above methods require a lot of perfect demonstrations to get satisfactory performance, which is different from POfD in this paper.<br />
<br />
For imitation learning, <br />
# Inverse Reinforce Learning [11] problems are solved by alternating between fitting the reward function and selecting the policy [12] [13]. But it cannot be extended to big-scale problems.<br />
# Generative Adversarial Imitation Learning (GAIL) [14] uses a discriminator to distinguish whether a state-action pair is from the expert or the learned policy and it can be applied into the high-dimensional continuous control problems.<br />
# An alternative imitation learning [26] is that an agent explores the environment without any expert supervision and distills this exploration data into goal-directed skills. These skills can then be used to imitate the visual demonstration provided by the expert.<br />
<br />
Both of the above methods are effective for imitation learning, but cannot leverage the valuable feedback given by the environments and usually suffer from bad performance when the expert data is imperfect. That is different from POfD in this paper.<br />
<br />
There is also another idea in which an agent learns using hybrid imitation learning and reinforcement learning reward[23, 24]. However, unlike this paper, they did not provide some theoretical support for their method and only explained some intuitive explanations.<br />
<br />
=Background=<br />
<br />
==Preliminaries==<br />
Markov Decision Process (MDP) [15] is defined by a tuple <math>⟨\mathcal{S}, \mathcal{A}, \mathcal{P}, r, \gamma⟩ </math>, where <math>\mathcal{S}</math> is the state space, <math>\mathcal{A} </math> is the action space, <math>\mathcal{P}(s'|s,a)</math> is the transition distribution of taking action <math> a </math> at state <math>s </math>, <math> r(s,a) </math>is the reward function, and <math> \gamma </math> is the discount factor between 0 and 1. Policy <math> \pi(a|s) </math> is a mapping from state to action probabilities, the performance of <math> \pi </math> is usually evaluated by its expected discounted reward <math> \eta(\pi) </math>: <br />
\[\eta(\pi)=\mathbb{E}_{\pi}[r(s,a)]=\mathbb{E}_{(s_0,a_0,s_1,...)}[\sum_{t=0}^\infty\gamma^{t}r(s_t,a_t)] \]<br />
The value function is <math> V_{\pi}(s) =\mathbb{E}_{\pi}[r(·,·)|s_0=s] </math>, the action value function is <math> Q_{\pi}(s,a) =\mathbb{E}_{\pi}[r(·,·)|s_0=s,a_0=a] </math>, and the advantage function that reflects the expected additional reward after taking action a at state s is <math> A_{\pi}(s,a)=Q_{\pi}(s,a)-V_{\pi}(s)</math>.<br />
Then the authors define Occupancy measure, which is used to estimate the probability that state <math>s</math> and state action pairs <math>(s,a)</math> when executing a certain policy.<br />
[[File:def1.png|500px|center]]<br />
Then the performance of <math> \pi </math> can be rewritten to: <br />
[[File:equ2.png|500px|center]]<br />
At the same time, the authors propose a lemma: <br />
[[File:lemma1.png|500px|center]]<br />
<br />
==Problem Definition==<br />
Generally, RL tasks and environments do not provide a comprehensive reward and instead rely on sparse feedback indicating whether the goal is reached.<br />
<br />
In this paper, the authors aim to develop a method that can boost exploration by leveraging effectively the demonstrations <math>D^E </math>from the expert policy <math> \pi_E </math> and maximize <math> \eta(\pi) </math> in the sparse-reward environment. The authors define the demonstrations <math>D^E=\{\tau_1,\tau_2,...,\tau_N\} </math>, where the i-th trajectory <math>\tau_i=\{(s_0^i,a_0^i),(s_1^i,a_1^i),...,(s_T^i,a_T^i)\} </math> is generated from the unknown expert policy <math>\pi_E </math>. In addition, there is an assumption on the quality of the expert policy:<br />
[[File:asp1.png|500px|center]]<br />
<br />
<br />
Throughout the paper, they use <math>\pi_E </math> to denote the expert policy that gives the relatively good <math>\eta_\pi </math>, and use <math>\hat{\mathbb{E}}_D </math>to denote empirical expectation estimated from the demonstrated trajectories <math>D^E </math>. We have the following reasonable and necessary assumption on the quality of the expert policy <math>\pi_E </math>.<br />
<br />
<br />
Moreover, it is not necessary to ensure that the expert policy is advantageous over all the policies. This is because that POfD will learn a better policy than expert policy by exploring on its own in later learning stages.<br />
<br />
=Method=<br />
<br />
==Policy Optimization with Demonstration (POfD)==<br />
<br />
[[File:ff1.png|thumb|500px|center |Figure 1: Demonstrations (the blue curve) enables POfD to explore in the high-reward regions (red arrows). On the other hand random explorations (olive green dashed curves) occur in sparse-reward environments.]]<br />
<br />
This method optimizes the policy by forcing the policy to explore in the nearby region of the expert policy that is specified by several demonstrated trajectories <math>D^E </math> (as shown in Fig.1) in order to avoid causing slow convergence or failure when the environment feedback is sparse. In addition, the authors encourage the policy π to explore by "following" the demonstrations <math>D^E </math>. Thus, a new learning objective is given:<br />
\[ \mathcal{L}(\pi_{\theta})=-\eta(\pi_{\theta})+\lambda_{1}D_{JS}(\pi_{\theta},\pi_{E})\]<br />
where <math>D_{JS}(\pi_{\theta},\pi_{E})</math> is Jensen-Shannon divergence between current policy <math>\pi_{\theta}</math> and the expert policy <math>\pi_{E}</math> , <math>\lambda_1</math> is a trading-off parameter, and <math>\theta</math> is policy parameter. According to Lemma 1, the authors use <math>D_{JS}(\rho_{\theta},\rho_{E})</math> to instead of <math>D_{JS}(\pi_{\theta},\pi_{E})</math>, because it is easier to optimize through adversarial training on demonstrations. The learning objective is: <br />
\[ \mathcal{L}(\pi_{\theta})=-\eta(\pi_{\theta})+\lambda_{1}D_{JS}(\rho_{\theta},\rho_{E})\]<br />
<br />
==Benefits of Exploration with Demonstrations==<br />
The authors introduce the benefits of POfD. Firstly, we consider the expression of expected return in policy gradient methods [16].<br />
\[ \eta(\pi)=\eta(\pi_{old})+\mathbb{E}_{\tau\sim\pi}[\sum_{t=0}^\infty\gamma^{t}A_{\pi_{old}}(s,a)]\]<br />
<math>\eta(\pi)</math>is the advantage over the policy <math>\pi_{old}</math> in the previous iteration, so the expression can be rewritten by<br />
\[ \eta(\pi)=\eta(\pi_{old})+\sum_{s}\rho_{\pi}(s)\sum_{a}\pi(a|s)A_{\pi_{old}}(s,a)\]<br />
The local approximation to <math>\eta(\pi)</math> up to first order is usually as the surrogate learning objective to be optimized by policy gradient methods due to the difficulties brought by complex dependency of <math>\rho_{\pi}(s)</math> over <math> \pi </math>:<br />
\[ J_{\pi_{old}}(\pi)=\eta(\pi_{old})+\sum_{s}\rho_{\pi_{old}}(s)\sum_{a}\pi(a|s)A_{\pi_{old}}(s,a)\]<br />
The policy gradient methods improve <math>\eta(\pi)</math> monotonically by optimizing the above <math>J_{\pi_{old}}(\pi)</math> with a sufficiently small update step from <math>\pi_{old}</math> to <math>\pi</math> such that <math>D_{KL}^{max}(\pi, \pi_{old})</math> is bounded [16] [17] [18]. POfD imposes an additional regularization <math>D_{JS}(\pi_{\theta}, \pi_{E})</math> between <math>\pi_\theta</math> and <math>\pi_{E}</math> in order to encourage explorations around regions demonstrated by the expert policy. Theorem 1 shows such benefits,<br />
[[File:them1.png|500px|center]]<br />
<br />
In fact, POfD brings another factor, <math>D_{J S}^{max}(\pi_{i}, \pi_{E})</math>, that would fully use the advantage <math>{\hat \delta}</math>and add improvements with a margin over pure policy gradient methods.<br />
<br />
==Optimization==<br />
<br />
For POfD, the authors choose to optimize the lower bound of the Jensen-Shannon divergence instead of directly optimizing the difficult Jensen-Shannon divergence. This optimization method is compatible with any policy gradient methods. Theorem 2 gives the lower bound of <math>D_{JS}(\rho_{\theta}, \rho_{E})</math>：<br />
[[File:them2.png|450px|center]]<br />
Thus, the occupancy measure matching objective can be written as:<br />
[[File:eqnlm.png|450px|center]]<br />
where <math> D(s,a)=\frac{1}{1+e^{-U(s,a)}}: \mathcal{S}\times \mathcal{A} \rightarrow (0,1)</math> is an arbitrary mapping function followed by a sigmoid activation function used for scaling, and its supremum ranging is like a discriminator for distinguishing whether the state-action pair is a current policy or an expert policy.<br />
To avoid overfitting, the authors add causal entropy <math>−H (\pi_{\theta}) </math> as the regularization term. Thus, the learning objective is: <br />
\[\min_{\theta}\mathcal{L}=-\eta(\pi_{\theta})-\lambda_{2}H(\pi_{\theta})+\lambda_{1} \sup_{{D\in(0,1)}^{S\times A}} \mathbb{E}_{\pi_{\theta}}[\log(D(s,a))]+\mathbb{E}_{\pi_{E}}[\log(1-D(s,a))]\]<br />
At this point, the problem closely resembles the minimax problem related to the Generative Adversarial Networks (GANs) [19]. The difference is that the discriminative model D of GANs is well-trained but the expert policy of POfD is not optimal. Then suppose D is parameterized by w. If it is from an expert policy, <math>D_w</math>is toward 1, otherwise it is toward 0. Thus, the minimax learning objective is:<br />
\[\min_{\theta}\max_{w}\mathcal{L}=-\eta(\pi_{\theta})-\lambda_{2}H (\pi_{\theta})+\lambda_{1}( \mathbb{E}_{\pi_{\theta}}[\log(D_{w}(s,a))]+\mathbb{E}_{\pi_{E}}[\log(1-D_{w}(s,a))])\]<br />
The minimax learning objective can be rewritten by substituting the expression of <math> \eta(\pi) </math>:<br />
\[\min_{\theta}\max_{w}-\mathbb{E}_{\pi_{\theta}}[r'(s,a)]-\lambda_{2}H (\pi_{\theta})+\lambda_{1}\mathbb{E}_{\pi_{E}}[\log(1-D_{w}(s,a))]\]<br />
where <math> r'(s,a)=r(a,b)-\lambda_{1}\log(D_{w}(s,a))</math> is the reshaped reward function.<br />
The above objective can be optimized efficiently by alternately updating policy parameters θ and discriminator parameters w, then the gradient is given by:<br />
\[\mathbb{E}_{\pi}[\nabla_{w}\log(D_{w}(s,a))]+\mathbb{E}_{\pi_{E}}[\nabla_{w}\log(1-D_{w}(s,a))]\]<br />
Then, fixing the discriminator <math>D_w</math>, the reshaped policy gradient is:<br />
\[\nabla_{\theta}\mathbb{E}_{\pi_{\theta}}[r'(s,a)]=\mathbb{E}_{\pi_{\theta}}[\nabla_{\theta}\log\pi_{\theta}(a|s)Q'(s,a)]\]<br />
where <math>Q'(\bar{s},\bar{a})=\mathbb{E}_{\pi_{\theta}}[r'(s,a)|s_0=\bar{s},a_0=\bar{a}]</math>.<br />
<br />
At the end, Algorithm 1 gives the detailed process.<br />
[[File:pofd.png|450px|center]]<br />
<br />
=Discussion on Existing LfD Methods=<br />
<br />
To connect with the proposed POfD method, interpretation of the existing methods DQfD and DDPGfD through occupancy measure matching is provided. Both of the existing methods leverage demonstrations to aid exploration in RL.<br />
<br />
==DQFD==<br />
DQFD [2] puts the demonstrations into a replay memory D and keeps them throughout the Q-learning process. The objective for DQFD is:<br />
\[J_{DQfD}={\hat{\mathbb{E}}}_{D}[(R_t(n)-Q_w(s_t,a_t))^2]+\alpha{\hat{\mathbb{E}}}_{D^E}[(R_t(n)-Q_w(s_t,a_t))^2]\]<br />
The second term can be rewritten as <math> {\hat{\mathbb{E}}}_{D^E}[(R_t(n)-Q_w(s_t,a_t))^2]={\hat{\mathbb{E}}}_{D^E}[(\hat{\rho}_E(s,a)-\rho_{\pi}(s,a))^{2}r^2(s,a)]</math>, which can be regarded as a regularization forcing current policy's occupancy measure to match the expert's empirical occupancy measure, weighted by the potential reward. Thus minimizing the objective<br />
with expert demonstration and self-generated off-policy datais actually equivalent to imposing an occupancy measure matching regularization to the original DQN objective.<br />
<br />
==DDPGfD==<br />
DDPGfD [3] also puts the demonstrations into a replay memory D, but it is based on an actor-critic framework [21]. The objective for DDPGfD is the same as DQFD. Its policy gradient is:<br />
\[\nabla_{\theta}J_{DDPGfD}\approx \mathbb{E}_{s,a}[\nabla_{a}Q_w(s,a)\nabla_{\theta}\pi_{\theta}(s)], a=\pi_{\theta}(s) \]<br />
From this equation, policy is updated relying on learned Q-network <math>Q_w </math>rather than the demonstrations <math>D^{E} </math>. DDPGfD shares the same objective function for <math>Q_w </math> as DQfD, thus they have the same way of leveraging demonstrations, that is the demonstrations in DQfD and DDPGfD induce an occupancy measure matching regularization.<br />
<br />
Although the above replay memory based LfD methods can benefit RL algorithms to some extent in sparse-reward environments, they have some limitations for sufficiently exploiting the demonstration data. First, such a paradigm utilizes expert trajectories only by treating them as learningreference, whose effect may be significantly underexploited when demonstrations are few, as indicated by the authors' experiments. Second, to be compatible with collected data during training, the demonstrated trajectories are required to be associated with rewards for each state transition. However, the rewards in demonstrations may differ from the ones used for learning the policy in the current environment [25], or they may be unavailable.<br />
<br />
=Experiments=<br />
<br />
==Goal==<br />
The authors aim at investigating 1) whether POfD can aid exploration by leveraging a few demonstrations, even though the demonstrations are imperfect. 2) whether POfD can succeed and achieve high empirical return, especially in environments where reward signals are sparse and rare. <br />
<br />
==Settings==<br />
The authors conduct the experiments on 8 physical control tasks, ranging from low-dimensional spaces to high-dimensional spaces and naturally sparse environments based on OpenAI Gym [20] and Mujoco (Multi-Joint dynamics with Contact) [5] (Gym is a toolkit for developing and comparing reinforcement learning algorithms. It supports teaching agents everything from walking to playing games like Pong or Pinball. MuJoCo is a physics engine aiming to facilitate research and development in robotics, biomechanics, graphics and animation, and other areas where fast and accurate simulation is needed. In order to get familiar with OpenAI Gym and Mujoco environment, you can watch these videos, respectively: [http://www.mujoco.org/image/home/mujocodemo.mp4 Mujoco], [https://gym.openai.com/v2018-02-21/videos/SpaceInvaders-v0-4184afb3-1223-4ac6-b52b-8e863cbe24a5/original.mp4 OpenAI Gym]). Due to the uniqueness of the environments, the authors introduce 4 ways to sparsify their built-in dense rewards. TYPE1: a reward of +1 is given when the agent reaches the terminal state, and otherwise 0. TYPE2: a reward of +1 is given when the agent survives for a while. TYPE3: a reward of +1 is given for every time the agent moves forward over a specific number of units in Mujoco environments. TYPE4: specially designed for InvertedDoublePendulum, a reward +1 is given when the second pole stays above a specific height of 0.89. The details are shown in Table 1. Moreover, only one single imperfect trajectory is used as the demonstrations in this paper. The authors collect the demonstrations by training an agent insufficiently by running TRPO (Trust Region Policy Optimization) in the corresponding dense environment. <br />
[[File:pofdt1.png|900px|center]]<br />
<br />
==Baselines==<br />
The authors compare POfD against 5 strong baselines:<br />
* training the policy with TRPO [17] in dense environments, which is called expert <br />
* training the policy with TRPO [17] in sparse environments<br />
* applying GAIL [14] to learn the policy from demonstrations<br />
* DQfD [2]<br />
* DDPGfD [3]<br />
<br />
<br />
1. Trust Region Policy Optimization (TRPO) is an iterative procedure for optimizing policies, with guaranteed monotonic improvement. By making several approximations to the theoretically-justified procedure, a practical algorithm such as this can be developed. This algorithm is similar to natural policy gradient methods and is effective for optimizing neural networks.<br />
<br />
2. Generative Adversarial Imitation Learning (GAIL) is a method to directly extract a policy from data as if it were obtained by reinforcement learning and by following inverse reinforcement learning.<br />
<br />
3. Deep Q-learning from Demonstrations (DQfD), is a method that leverages small sets of demonstration data to speed up the learning process from relatively small amounts of demonstration data and is able to automatically assess the necessary ratio of demonstration data while learning thanks to a prioritized replay mechanism.<br />
<br />
4. DDPGfD (Deep Deterministic Policy Gradients From Demonstrations) uses prioritized replay to enable efficient propagation of the reward information, which is essential in problems with sparse rewards.<br />
<br />
==Results==<br />
Firstly, the authors test the performance of POfD in sparse control environments with discrete actions. From Table 1, POfD achieves performance comparable with the policy learned under dense environments. From Figure 2, only POfD successes to explore sufficiently and achieves great performance in both sparse environments. TRPO [17] and DQFD [2] fail to explore and GAIL [14] converges to the imperfect demonstration in MountainCar [22].<br />
<br />
[[File:pofdf2.png|500px|center]]<br />
<br />
Then, the authors test the performance of POfD under spares environments with continuous actions space. From Figure 3, POfD achieves expert-level performance in terms of accumulated rewards and surpasses other strong baselines training the policy with TRPO. By watching the learning process of different methods, we can see that TRPO consistently fails to explore the environments when the feedback is sparse, except for HalfCheetah. This may be because there is no terminal state in HalfCheetah, thus a random agent can perform reasonably well as long as the time horizon is sufficiently long. This is shown in Figure3 where the improvement of TRPO begins to show after 400 iterations. DDPGfD and GAIL have common drawback: during training process, they both converge to the imperfect demonstration data. For HalfCheetah, GAIL fails to converge and DDPGfD converges to an even worse point. This situation is expected because the policy and value networks tend to over-fit when having few data, so the training process of GAIL and DDPGfD is severely biased by the imperfect data. Finally, our proposed method can effectively explore the environment with the help of demonstration-based intrinsic reward reshaping and succeeds consistently across different tasks both in terms of learning stability and convergence speed.<br />
[[File:pofdf3.png|900px|center]]<br />
<br />
The authors also implement a locomotion task <math>Humanoid</math>, which teaches a human-like robot to walk. The state space of dimension is 376, which is very hard to render. As a result, POfD still outperformed all three baselike methods, as they failed to learn policies in such a sparse reward environment.<br />
<br />
The reacher environment is a task that the target is to control a robot arm to touch an object. the location of the object is random for each instantiation. The environment reward is sparse: every time the arm reaches the ball and holds for a while (e.g., 5 time steps), it receives a reward of +1; otherwise, it gets zero reward. The authors select 15 random trajectories as demonstration data, and the performance of POfD is much better than the expert, while all other baseline methods failed.<br />
<br />
=Conclusion=<br />
In this paper, POfD is proposed that acquires knowledge from a limited amount of imperfect demonstration data to aid exploration in environments with sparse feedback. It is compatible with any policy gradient method. POfD induces implicit dynamic reward shaping and brings provable benefits for policy improvement. Moreover, the results of the experiments have shown the validity and effectiveness of POfD in encouraging the agent to explore around the nearby region of the expert policy and learn better policies. The key contribution is that POfD helps the agent work with few and imperfect demonstrations in an environment with sparse rewards.<br />
<br />
=Critique=<br />
# A novel demonstration-based policy optimization method is proposed. In the process of policy optimization, POfD reshapes the reward function. This new reward function can guide the agent to imitate the expert behavior when the reward is sparse and explore on its own when the reward value can be obtained, which can take full advantage of the demonstration data and there is no need to ensure that the expert policy is the optimal policy.<br />
# POfD can be combined with any policy gradient methods. Its performance surpasses five strong baselines and can be comparable to the agents trained in the dense-reward environment.<br />
# The paper is structured and the flow of ideas is easy to follow. For related work, the authors clearly explain similarities and differences among these related works.<br />
# This paper's scalability is demonstrated. The experiments environments are ranging from low-dimensional spaces to high-dimensional spaces and from discrete action spaces to continuous actions spaces. For future work, can it be realized in the real world?<br />
# There is a doubt that whether it is a correct method to use the trajectory that was insufficiently learned in a dense-reward environment as the imperfect demonstration.<br />
# In this paper, the performance only is judged by the cumulative reward, can other evaluation terms be considered? For example, the convergence rate.<br />
# The performance of this algorithm hinges on the assumption that expert demonstrations are near optimal in the action space. As seen in figure 3, there appears to be an upper bound to performance near (or just above) the expert accuracy -- this may be an indication of a performance ceiling. In games where near-optimal policies can differ greatly (e.g.; offensive or defensive strategies in chess), the success of the model will depend on the selection of expert demonstrations that are closest to a truly optimal policy (i.e.; just because a policy is the current expert, it does not mean it resembles the true optimal policy).<br />
<br />
=References=<br />
[1] Nair, A., McGrew, B., Andrychowicz, M., Zaremba, W., and Abbeel, P. Overcoming exploration in reinforcement learning with demonstrations. arXiv preprint arXiv:1709.10089, 2017.<br />
<br />
[2] Hester, T., Vecerik, M., Pietquin, O., Lanctot, M., Schaul, T., Piot, B., Sendonaris, A., Dulac-Arnold, G., Osband, I., Agapiou, J., et al. Learning from demonstrations for real world reinforcement learning. arXiv preprint arXiv:1704.03732, 2017.<br />
<br />
[3] Večerík, M., Hester, T., Scholz, J., Wang, F., Pietquin, O., Piot, B., Heess, N., Rotho ̈rl, T., Lampe, T., and Riedmiller, M. Leveraging demonstrations for deep reinforcement learning on robotics problems with sparse rewards. arXiv preprint arXiv:1707.08817, 2017.<br />
<br />
[4] Silver, D., Huang, A., Maddison, C. J., Guez, A., Sifre, L., Van Den Driessche, G., Schrittwieser, J., Antonoglou, I., Panneershelvam, V., Lanctot, M., et al. Mastering the game of go with deep neural networks and tree search. nature, 529(7587):484–489, 2016.<br />
<br />
[5] Todorov, E., Erez, T., and Tassa, Y. Mujoco: A physics engine for model-based control. In Intelligent Robots and Systems (IROS), 2012 IEEE/RSJ International Con- ference on, pp. 5026–5033. IEEE, 2012.<br />
<br />
[6] Schaal, S. Learning from demonstration. In Advances in neural information processing systems, pp. 1040–1046, 1997.<br />
<br />
[7] Kim, B., Farahmand, A.-m., Pineau, J., and Precup, D. Learning from limited demonstrations. In Advances in Neural Information Processing Systems, pp. 2859–2867, 2013.<br />
<br />
[8] Piot, B., Geist, M., and Pietquin, O. Boosted bellman resid- ual minimization handling expert demonstrations. In Joint European Conference on Machine Learning and Knowl- edge Discovery in Databases, pp. 549–564. Springer, 2014.<br />
<br />
[9] Aravind S. Lakshminarayanan, Sherjil Ozair, Y. B. Rein- forcement learning with few expert demonstrations. In NIPS workshop, 2016.<br />
<br />
[10] Brys, T., Harutyunyan, A., Suay, H. B., Chernova, S., Tay- lor, M. E., and Nowe ́, A. Reinforcement learning from demonstration through shaping. In IJCAI, pp. 3352–3358, 2015.<br />
<br />
[11] Ng, A. Y., Russell, S. J., et al. Algorithms for inverse reinforcement learning. In Icml, pp. 663–670, 2000.<br />
<br />
[12] Syed, U. and Schapire, R. E. A game-theoretic approach to apprenticeship learning. In Advances in neural informa- tion processing systems, pp. 1449–1456, 2008.<br />
<br />
[13] Syed, U., Bowling, M., and Schapire, R. E. Apprenticeship learning using linear programming. In Proceedings of the 25th international conference on Machine learning, pp. 1032–1039. ACM, 2008.<br />
<br />
[14] Ho, J. and Ermon, S. Generative adversarial imitation learn- ing. In Advances in Neural Information Processing Sys- tems, pp. 4565–4573, 2016.<br />
<br />
[15] Sutton, R. S. and Barto, A. G. Reinforcement learning: An introduction, volume 1. MIT press Cambridge, 1998.<br />
<br />
[16] Kakade, S. M. A natural policy gradient. In Advances in neural information processing systems, pp. 1531–1538, 2002.<br />
<br />
[17] Schulman, J., Levine, S., Abbeel, P., Jordan, M., and Moritz, P. Trust region policy optimization. In Proceedings of the 32nd International Conference on Machine Learning (ICML-15), pp. 1889–1897, 2015.<br />
<br />
[18] Schulman, J., Wolski, F., Dhariwal, P., Radford, A., and Klimov, O. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347, 2017.<br />
<br />
[19] Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., and Bengio, Y. Generative adversarial nets. In Advances in neural information processing systems, pp. 2672–2680, 2014.<br />
<br />
[20] Brockman, G., Cheung, V., Pettersson, L., Schneider, J., Schulman, J., Tang, J., and Zaremba, W. Openai gym, 2016.<br />
<br />
[21] Lillicrap, T. P., Hunt, J. J., Pritzel, A., Heess, N., Erez, T., Tassa, Y., Silver, D., and Wierstra, D. Continuous control with deep reinforcement learning. arXiv preprint arXiv:1509.02971, 2015.<br />
<br />
[22] Moore, A. W. Efficient memory-based learning for robot control. 1990.<br />
<br />
[23] Zhu, Y., Wang, Z., Merel, J., Rusu, A., Erez, T., Cabi, S., Tunyasuvunakool, S., Kramar, J., Hadsell, R., de Freitas, N., et al. Reinforcement and imitation learning for diverse visuomotor skills. arXiv preprint arXiv:1802.09564, 2018.<br />
<br />
[24] Li, Y., Song, J., and Ermon, S. Infogail: Interpretable imitation learning from visual demonstrations. In Advances in Neural Information Processing Systems, pp. 3815–3825, 2017.<br />
<br />
[25] Ziebart, B. D., Maas, A. L., Bagnell, J. A., and Dey, A. K. Maximum entropy inverse reinforcement learning. In AAAI, volume 8, pp. 1433–1438. Chicago, IL, USA, 2008.<br />
<br />
[26] Pathak, D., Mahmoudieh, P., Luo, G., Agrawal, P., Chen, D., Shentu, Y., Shelhamer, E., Malik, J., Efros, A. A., and Darrell, T. Zero-Shot Visual Imitation. In International Conference on Learning Representations (ICLR), 2018.</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=ShakeDrop_Regularization&diff=42350ShakeDrop Regularization2018-12-08T06:41:50Z<p>Gchalato: /* Existing Methods */</p>
<hr />
<div>=Introduction=<br />
Current state of the art techniques for object classification are deep neural networks based on the residual block, first published by (He et al., 2016). This technique has been the foundation of several improved networks, including Wide ResNet (Zagoruyko & Komodakis, 2016), PyramdNet (Han et al., 2017) and ResNeXt (Xie et al., 2017). They have been further improved by regularization, such as Stochastic Depth (ResDrop) (Huang et al., 2016) and Shake-Shake (Gastaldi, 2017), which can avoid some problem like vanishing gradients. Shake-Shake applied to ResNeXt has achieved one of the lowest error rates on the CIFAR-10 and CIFAR-100 datasets. However, it is only applicable to multi-branch architectures and is not memory efficient since it requires two branches of residual blocks to apply. Note that the authors of Shake-Shake are rejecting the claim of their memory inefficiency. They claimed that there is no memory issue, just because there are <math>2\times</math> branches doesn't mean Shake-Shake needs <math>2\times</math> memory as it can use less memory to achieve the same performance.<br />
<br />
To address this problem, ShakeDrop regularization that can realize a similar disturbance to Shake-Shake on a single residual block is proposed.ShakeDrop disturbs learning more strongly by multiplying even a negative factor to the output of a convolutional layer in the forward training pass. In addition, a different factor from the forward pass is multiplied in the backward training pass. As a byproduct, however, learning process gets unstable. Moreover, they use ResDrop to stabilize the learning process. This paper seeks to formulate a general expansion of Shake-Shake that can be applied to any residual block based network.<br />
<br />
=Existing Methods=<br />
<br />
'''Deep Approaches'''<br />
<br />
'''ResNet''', was the first use of residual blocks, a foundational feature in many modern state of the art convolution neural networks. They can be formulated as <math>G(x) = x + F(x)</math> where <math>x</math> and <math>G(x)</math> are the input and output of the residual block, and <math>F(x)</math> is the output of the residual branch on the residual block. A residual block typically performs a convolution operation and then passes the result plus its input onto the next block. <br />
<br />
Intuition behind Residual blocks:<br />
If the identity mapping is optimal, We can easily push the residuals to zero (F(x) = 0) than to fit an identity mapping (x, input=output) by a stack of non-linear layers. In simple language it is very easy to come up with a solution like F(x) =0 rather than F(x)=x using stack of non-linear cnn layers as function (Think about it). So, this function F(x) is what the authors called Residual function ([https://medium.com/@14prakash/understanding-and-implementing-architectures-of-resnet-and-resnext-for-state-of-the-art-image-cf51669e1624 Reference]).<br />
<br />
Residual blocks are used for two main reasons. First, as our networks become “deeper” and more flexible, we also need to take many more gradients during backpropogation. This exponentially increases the risk of vanishing gradients, particularly with state-of-the art structures. To counter this, residual layers pass entire layers – with the identity function applied – further down the network. Intuitively, this gives gives higher gradient values. Secondly, this gives the network another path to work on. If forced non-linearity is not an optimal choice, the network can bypass it through these residual blocks.<br />
<br />
<br />
[[File:ResidualBlock.png|580px|centre|thumb|An example of a simple residual block from Deep Residual Learning for Image Recognition by He et al., 2016]]<br />
<br />
ResNet is constructed out of a large number of these residual blocks sequentially stacked. It is interesting to note that having too many layers can cause overfitting, as pointed out by He et al. (2016) with the high error rates for the 1,202-layer ResNet on CIFAR datasets. Another paper (Veit et al., 2016) empirically showed that the cause of the high error rates can be mostly attributed to specific residual blocks whose channels increase greatly.<br />
<br />
'''PyramidNet''' is an important iteration that built on ResNet and WideResNet by gradually increasing channels on each residual block. The residual block is similar to those used in ResNet. It has been used to generate some of the first successful convolution neural networks with very large depth, at 272 layers. Amongst unmodified residual network architectures, it performs the best on the CIFAR datasets.<br />
<br />
[[File:ResidualBlockComparison.png|980px|centre|thumb|A simple illustration of different residual blocks from Deep Pyramidal Residual Networks by Han et al., 2017. The width of a block reflects the number of channels used in that layer.]]<br />
<br />
<br />
'''Non-Deep Approaches'''<br />
<br />
'''Wide ResNet''' modified ResNet by increasing channels in each layer, having a wider and shallower structure. Similarly to PyramidNet, this architecture avoids some of the pitfalls in the original formulation of ResNet.<br />
<br />
'''ResNeXt''' achieved performance beyond that of Wide ResNet with only a small increase in the number of parameters. It can be formulated as <math>G(x) = x + F_1(x)+F_2(x)</math>. In this case, <math>F_1(x)</math> and <math>F_2(x)</math> are the outputs of two paired convolution operations in a single residual block. The number of branches is not limited to 2, and will control the result of this network.<br />
<br />
<br />
[[File:SimplifiedResNeXt.png|600px|centre|thumb|Simplified ResNeXt Convolution Block. Yamada et al., 2018]]<br />
<br />
<br />
'''Regularization Methods For Residual Blocks'''<br />
<br />
'''Stochastic Depth''' works by randomly dropping paths in the residual blocks. On the <math>l^{th}</math> residual block the Stochastic Depth process is given as <math>G(x)=x+b_lF(x)</math> where <math>b_l \in \{0,1\}</math> is a Bernoulli random variable with probability <math>p_l</math>. Unlike sequential networks, there are many paths from the input to the output in these networks. By dropping some of the connections, the network is forced to flow through different paths to get the final deep layer representation. In a way it is similar to dropout, but for paths in multi-path networks. Using a constant value for <math>p_l</math> didn't work well, so instead a linear decay rule <math>p_l = 1 - \frac{l}{L}(1-p_L)</math> was used. In this equation, <math>L</math> is the number of layers, and <math>p_L</math> is the initial parameter. Essentially, the probability of a connection dropping in inversely proportional to the its depth in the network.<br />
<br />
'''Shake-Shake''' is a regularization method that specifically improves the ResNeXt (multiple residual connections) architecture. It is given as <math>G(x)=x+\alpha F_1(x)+(1-\alpha)F_2(x)</math>, where <math>\alpha \in [0,1]</math> is a random coefficient. Essentially, one of the parallel residual connections is dropped in the forward direction. This is similar to stochastic depth regularization, but a residual path always exists.<br />
Moreover, on the backward pass a similar random variable <math>\beta</math> is used to independently drop paths for gradient flow. This has the effect of adding noise in the gradients update process and improved performance over the vanilla ResNeXt network.<br />
<br />
<br />
[[File:Paper 32.jpg|600px|centre|thumb| Shake-Shake (ResNeXt + Shake-Shake) (Gastaldi, 2017), in which some processing layers omitted for conciseness.]]<br />
<br />
=Proposed Method=<br />
We give an intuitive interpretation of the forward pass of Shake-Shake regularization. To the best of our knowledge, it has not been given yet, while the phenomenon in the backward pass is experimentally investigated by Gastaldi (2017). In the forward pass, Shake-Shake interpolates the outputs of two residual branches with a random variable α that controls the degree of interpolation. As DeVries & Taylor (2017a) demonstrated that interpolation of two data in the feature space can synthesize reasonable augmented data, the interpolation of two residual blocks of Shake-Shake in the forward pass can be interpreted as synthesizing data. Use of a random variable α generates many different augmented data. On the other hand, in the backward pass, a different random variable β is used to disturb learning to make the network learnable long time. Gastaldi (2017) demonstrated how the difference between <math>\alpha</math> and <math>\beta</math> affects.<br />
<br />
The regularization mechanism of Shake-Shake relies on two or more residual branches, so that it can be applied only to 2-branch networks architectures. In addition, 2-branch network architectures consume more memory than 1-branch network architectures. One may think the number of learnable parameters of ResNeXt can be kept in 1-branch and 2-branch network architectures by controlling its cardinality and the number of channels (filters). For example, a 1-branch network (e.g., ResNeXt 1-64d) and its corresponding 2-branch network (e.g., ResNeXt 2-40d) have almost same number of learnable parameters. However, even so, it increases memory consumption due to the overhead to keep the inputs of residual blocks and so on. By comparing ResNeXt 1-64d and 2-40d, the latter requires more memory than the former by 8% in theory (for one layer) and by 11% in measured values (for 152 layers).<br />
<br />
This paper seeks to generalize the method proposed in Shake-Shake to be applied to any residual structure network. Shake-Shake. The initial formulation of 1-branch shake is <math>G(x) = x + \alpha F(x)</math>. In this case, <math>\alpha</math> is a coefficient that disturbs the forward pass, but is not necessarily constrained to be [0,1]. Another corresponding coefficient <math>\beta</math> is used in the backwards pass. Applying this simple adaptation of Shake-Shake on a 110-layer version of PyramidNet with <math>\alpha \in [0,1]</math> and <math>\beta \in [0,1]</math> performs abysmally, with an error rate of 77.99%.<br />
<br />
This failure is a result of the setup causing too much perturbation. A trick is needed to promote learning with large perturbations, to preserve the regularization effect. The idea of the authors is to borrow from ResDrop and combine that with Shake-Shake. This works by randomly deciding whether to apply 1-branch shake. This creates in effect two networks, the original network without a regularization component, and a regularized network. When mixing up two networks, we expected the following effects: When the non regularized network is selected, learning is promoted; when the perturbed network is selected, learning is disturbed. Achieving good performance requires a balance between the two. <br />
<br />
'''ShakeDrop''' is given as <br />
<br />
<div align="center"><br />
<math>G(x) = x + (b_l + \alpha - b_l \alpha)F(x)</math>,<br />
</div><br />
<br />
where <math>b_l</math> is a Bernoulli random variable following the linear decay rule used in Stochastic Depth. An alternative presentation is <br />
<br />
<div align="center"><br />
<math><br />
G(x) = \begin{cases}<br />
x + F(x) ~~ \text{if } b_l = 1 \\<br />
x + \alpha F(x) ~~ \text{otherwise}<br />
\end{cases}<br />
</math><br />
</div><br />
<br />
If <math>b_l = 1</math> then ShakeDrop is equivalent to the original network, otherwise it is the network + 1-branch Shake. The authors also found that the linear decay rule of ResDrop works well, compared with the uniform rule. Regardless of the value of <math>\beta</math> on the backwards pass, network weights will be updated.<br />
<br />
=Experiments=<br />
<br />
'''Parameter Search'''<br />
<br />
The authors experiments began with a hyperparameter search utilizing ShakeDrop on pyramidal networks. The PyramidNet used was made up of a total of 110 layers which included a convolutional layer and a final fully connected layer. It had 54 additive pyramidal residual blocks and the final residual block had 286 channels. The results of this search are presented below. <br />
<br />
[[File:ShakeDropHyperParameterSearch.png|600px|centre|thumb|Average Top-1 errors (%) of “PyramidNet + ShakeDrop” with several ranges of parameters of 4 runs at the final (300th) epoch on CIFAR-100 dataset in the “Batch” level. In some settings, it is equivalent to PyramidNet and PyramidDrop. Borrowed from ShakeDrop Regularization by Yamada et al., 2018.]]<br />
<br />
The setting that are used throughout the rest of the experiments are then <math>\alpha \in [-1,1]</math> and <math>\beta \in [0,1]</math>. Cases H and F outperform PyramidNet, suggesting that the strong perturbations imposed by ShakeDrop are functioning as intended. However, fully applying the perturbations in the backwards pass appears to destabilize the network, resulting in performance that is worse than standard PyramidNet.<br />
<br />
[[File:ParameterUpdateShakeDrop.png|400px|centre]]<br />
<br />
Following this initial parameter decision, the authors tested 4 different strategies for parameter update among "Batch" (same coefficients for all images in minibatch for each residual block), "Image" (same scaling coefficients for each image for each residual block), "Channel" (same scaling coefficients for each element for each residual block), and "Pixel" (same scaling coefficients for each element for each residual block). While Pixel was the best in terms of error rate, it is not very memory efficient, so Image was selected as it had the second best performance without the memory drawback.<br />
<br />
'''Comparison with Regularization Methods'''<br />
<br />
For these experiments, there are a few modifications that were made to assist with training. For ResNeXt, the EraseRelu formulation has each residual block ends in batch normalization. The Wide ResNet also is compared between vanilla with batch normalization and without. Batch normalization keeps the outputs of residual blocks in a certain range, as otherwise <math>\alpha</math> and <math>\beta</math> could cause perturbations that are too large, causing divergent learning. There is also a comparison of ResDrop/ShakeDrop Type A (where the regularization unit is inserted before the add unit for a residual branch) and after (where the regularization unit is inserted after the add unit for a residual branch). <br />
<br />
These experiments are performed on the CIFAR-100 dataset.<br />
<br />
[[File:ShakeDropArchitectureComparison1.png|800px|centre|thumb|]]<br />
<br />
[[File:ShakeDropArchitectureComparison2.png|800px|centre|thumb|]]<br />
<br />
[[File:ShakeDropArchitectureComparison3.png|800px|centre|thumb|]]<br />
<br />
For a final round of testing, the training setup was modified to incorporate other techniques used in state of the art methods. For most of the tests, the learning rate for the 300 epoch version started at 0.1 and decayed by a factor of 0.1 1/2 & 3/4 of the way through training. The alternative was cosine annealing, based on the presentation by Loshchilov and Hutter in their paper SGDR: Stochastic Gradient Descent with Warm Restarts. This is indicated in the Cos column, with a check indicating cosine annealing. <br />
<br />
[[File:CosineAnnealing.png|400px|centre|thumb|]]<br />
<br />
The Reg column indicates the regularization method used, either none, ResDrop (RD), Shake-Shake (SS), or ShakeDrop (SD). Fianlly, the Fil Column determines the type of data augmentation used, either none, cutout (CO) (DeVries & Taylor, 2017b), or Random Erasing (RE) (Zhong et al., 2017). <br />
<br />
[[File:ShakeDropComparison.png|800px|centre|thumb|Top-1 Errors (%) at final epoch on CIFAR-10/100 datasets]]<br />
<br />
'''State-of-the-Art Comparisons'''<br />
<br />
A direct comparison with state of the art methods is favorable for this new method. <br />
<br />
# Fair comparison of ResNeXt + Shake-Shake with PyramidNet + ShakeDrop gives an improvement of 0.19% on CIFAR-10 and 1.86% on CIFAR-100. Under these conditions, the final error rate is then 2.67% for CIFAR-10 and 13.99% for CIFAR-100.<br />
# Fair comparison of ResNeXt + Shake-Shake + Cutout with PyramidNet + ShakeDrop + Random Erasing gives an improvement of 0.25% on CIFAR-10 and 3.01% on CIFAR 100. Under these conditions, the final error rate is then 2.31% for CIFAR-10 and 12.19% for CIFAR-100.<br />
# Comparison with the state-of-the-arts, PyramidNet + ShakeDrop gives an improvement of 0.25% on CIFAR-10 than ResNeXt + Shake-Shake + Cutout, PyramidNet + ShakeDrop gives an improvement of 2.85% on CIFAR-100 than Coupled Ensemble.<br />
<br />
=Implementation details=<br />
<br />
'''CIFAR-10/100 datasets'''<br />
<br />
All the images in these datasets were color normalized and then horizontally flipped with a probability of 50%. All of then then were zero padded to have a dimentionality of 40 by 40 pixels.<br />
<br />
<br />
=Conclusion=<br />
The paper proposes a new form of regularization that is an extension of "Shake-Shake" regularization [Gastaldi, 2017]. The original "shake-shake" proposes using two residual paths adding to the same output, and during training, considering different randomly selected convex combinations of the two paths (while using an equally weighted combination at test time). This paper contends that this requires additional memory, and attempts to achieve similar regularization with a single path. To do so, they train a network with a single residual path, where the residual is included without attenuation in some cases with some fixed probability, and attenuated randomly (or even inverted) in others. The paper contends that this achieves superior performance than choosing simply a random attenuation for every sample (although, this can be seen as choosing an attenuation under a distribution with some fixed probability mass.<br />
<br />
Their stochastic regularization method, ShakeDrop, which outperforms previous state of the art methods while maintaining similar memory efficiency. It demonstrates that heavily perturbing a network can help to overcome issues with overfitting. It is also an effective way to regularize residual networks for image classification. The method was tested by CIFAR-10/100 and Tiny ImageNet datasets and showed great performance.<br />
<br />
=Critique=<br />
<br />
The novelty of this paper is low as pointed out by the reviewers. Also, there is a confusion whether or not the results could be replicated as <math>\alpha</math> and <math>\beta</math> are choosen randomly. The proposed ShakeDrop regularization is essentially a combination of the PyramidDrop and Shake-Shake regularization. The most surprising part is that the forward weight can be negative thus inverting the output of a convolution. The mathematical justification for ShakeDrop regularization is limited, relying on intuition and empirical evidence instead.<br />
<br />
One downside of this methods (as was identified in the presentation as well) is that the training for cosine annealing variation of the model takes 1800 epochs which is time intensive compared to other methods that were compared as baselines. This can limit practical implementation of this algorithm.<br />
<br />
As pointed out from the above, the method basically relies heavily on the intuition. This means that the performance of the algorithm can not been extended beyond the CIFAR dataset and can vary a lot depending on the characteristics of data sets that users are performing, with some exaggeration. However, the performance is still impressive since it performs better than known algorithms. It is not clear as to how the proposed technique would work with a non-residual architecture.<br />
It lacks conclusive proof that "shake-drop" is a generically useful regularization technique. For one, the method is evaluated only on small toy-datasets: CIFAR-10 and CIFAR-100. Evaluation on Imagenet perhaps would have been valuable. There is also another dataset that would of been good to try SVHN. Overall I believe the impact of this beyond CIFAR is unclear.<br />
<br />
=References=<br />
[Yamada et al., 2018] Yamada Y, Iwamura M, Kise K. ShakeDrop regularization. arXiv preprint arXiv:1802.02375. 2018 Feb 7.<br />
<br />
[He et al., 2016] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proc. CVPR, 2016.<br />
<br />
[Zagoruyko & Komodakis, 2016] Sergey Zagoruyko and Nikos Komodakis. Wide residual networks. In Proc. BMVC, 2016.<br />
<br />
[Han et al., 2017] Dongyoon Han, Jiwhan Kim, and Junmo Kim. Deep pyramidal residual networks. In Proc. CVPR, 2017a.<br />
<br />
[Xie et al., 2017] Saining Xie, Ross Girshick, Piotr Dollar, Zhuowen Tu, and Kaiming He. Aggregated residual transformations for deep neural networks. In Proc. CVPR, 2017.<br />
<br />
[Huang et al., 2016] Gao Huang, Yu Sun, Zhuang Liu, Daniel Sedra, and Kilian Weinberger. Deep networks with stochastic depth. arXiv preprint arXiv:1603.09382v3, 2016.<br />
<br />
[Gastaldi, 2017] Xavier Gastaldi. Shake-shake regularization. arXiv preprint arXiv:1705.07485v2, 2017.<br />
<br />
[Loshilov & Hutter, 2016] Ilya Loshchilov and Frank Hutter. Sgdr: Stochastic gradient descent with warm restarts. arXiv preprint arXiv:1608.03983, 2016.<br />
<br />
[DeVries & Taylor, 2017b] Terrance DeVries and Graham W. Taylor. Improved regularization of convolutional neural networks with cutout. arXiv preprint arXiv:1708.04552, 2017b.<br />
<br />
[Zhong et al., 2017] Zhun Zhong, Liang Zheng, Guoliang Kang, Shaozi Li, and Yi Yang. Random erasing data augmentation. arXiv preprint arXiv:1708.04896, 2017.<br />
<br />
[Dutt et al., 2017] Anuvabh Dutt, Denis Pellerin, and Georges Qunot. Coupled ensembles of neural networks. arXiv preprint 1709.06053v1, 2017.<br />
<br />
[Veit et al., 2016] Andreas Veit, Michael J Wilber, and Serge Belongie. Residual networks behave like ensembles of relatively shallow networks. Advances in Neural Information Processing Systems 29, 2016.</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=ShakeDrop_Regularization&diff=42349ShakeDrop Regularization2018-12-08T06:41:20Z<p>Gchalato: /* Existing Methods */</p>
<hr />
<div>=Introduction=<br />
Current state of the art techniques for object classification are deep neural networks based on the residual block, first published by (He et al., 2016). This technique has been the foundation of several improved networks, including Wide ResNet (Zagoruyko & Komodakis, 2016), PyramdNet (Han et al., 2017) and ResNeXt (Xie et al., 2017). They have been further improved by regularization, such as Stochastic Depth (ResDrop) (Huang et al., 2016) and Shake-Shake (Gastaldi, 2017), which can avoid some problem like vanishing gradients. Shake-Shake applied to ResNeXt has achieved one of the lowest error rates on the CIFAR-10 and CIFAR-100 datasets. However, it is only applicable to multi-branch architectures and is not memory efficient since it requires two branches of residual blocks to apply. Note that the authors of Shake-Shake are rejecting the claim of their memory inefficiency. They claimed that there is no memory issue, just because there are <math>2\times</math> branches doesn't mean Shake-Shake needs <math>2\times</math> memory as it can use less memory to achieve the same performance.<br />
<br />
To address this problem, ShakeDrop regularization that can realize a similar disturbance to Shake-Shake on a single residual block is proposed.ShakeDrop disturbs learning more strongly by multiplying even a negative factor to the output of a convolutional layer in the forward training pass. In addition, a different factor from the forward pass is multiplied in the backward training pass. As a byproduct, however, learning process gets unstable. Moreover, they use ResDrop to stabilize the learning process. This paper seeks to formulate a general expansion of Shake-Shake that can be applied to any residual block based network.<br />
<br />
=Existing Methods=<br />
<br />
'''Deep Approaches'''<br />
<br />
'''ResNet''', was the first use of residual blocks, a foundational feature in many modern state of the art convolution neural networks. They can be formulated as <math>G(x) = x + F(x)</math> where <math>x</math> and <math>G(x)</math> are the input and output of the residual block, and <math>F(x)</math> is the output of the residual branch on the residual block. A residual block typically performs a convolution operation and then passes the result plus its input onto the next block. <br />
<br />
Intuition behind Residual blocks:<br />
If the identity mapping is optimal, We can easily push the residuals to zero (F(x) = 0) than to fit an identity mapping (x, input=output) by a stack of non-linear layers. In simple language it is very easy to come up with a solution like F(x) =0 rather than F(x)=x using stack of non-linear cnn layers as function (Think about it). So, this function F(x) is what the authors called Residual function ([https://medium.com/@14prakash/understanding-and-implementing-architectures-of-resnet-and-resnext-for-state-of-the-art-image-cf51669e1624 Reference]).<br />
<br />
Residual blocks are used for two main reasons. First, as our networks become “deeper” and more flexible, we also need to take many more gradients during backpropogation. This exponentially increases the risk of vanishing gradients, particularly with state-of-the art structures. To counter this, residual layers use the identity function to pass entire layers – with the identity function applied – further down the network. Intuitively, this gives gives higher gradient values. Secondly, this gives the network another path to work on. If forced non-linearity is not an optimal choice, the network can bypass it through these residual blocks.<br />
<br />
<br />
[[File:ResidualBlock.png|580px|centre|thumb|An example of a simple residual block from Deep Residual Learning for Image Recognition by He et al., 2016]]<br />
<br />
ResNet is constructed out of a large number of these residual blocks sequentially stacked. It is interesting to note that having too many layers can cause overfitting, as pointed out by He et al. (2016) with the high error rates for the 1,202-layer ResNet on CIFAR datasets. Another paper (Veit et al., 2016) empirically showed that the cause of the high error rates can be mostly attributed to specific residual blocks whose channels increase greatly.<br />
<br />
'''PyramidNet''' is an important iteration that built on ResNet and WideResNet by gradually increasing channels on each residual block. The residual block is similar to those used in ResNet. It has been used to generate some of the first successful convolution neural networks with very large depth, at 272 layers. Amongst unmodified residual network architectures, it performs the best on the CIFAR datasets.<br />
<br />
[[File:ResidualBlockComparison.png|980px|centre|thumb|A simple illustration of different residual blocks from Deep Pyramidal Residual Networks by Han et al., 2017. The width of a block reflects the number of channels used in that layer.]]<br />
<br />
<br />
'''Non-Deep Approaches'''<br />
<br />
'''Wide ResNet''' modified ResNet by increasing channels in each layer, having a wider and shallower structure. Similarly to PyramidNet, this architecture avoids some of the pitfalls in the original formulation of ResNet.<br />
<br />
'''ResNeXt''' achieved performance beyond that of Wide ResNet with only a small increase in the number of parameters. It can be formulated as <math>G(x) = x + F_1(x)+F_2(x)</math>. In this case, <math>F_1(x)</math> and <math>F_2(x)</math> are the outputs of two paired convolution operations in a single residual block. The number of branches is not limited to 2, and will control the result of this network.<br />
<br />
<br />
[[File:SimplifiedResNeXt.png|600px|centre|thumb|Simplified ResNeXt Convolution Block. Yamada et al., 2018]]<br />
<br />
<br />
'''Regularization Methods For Residual Blocks'''<br />
<br />
'''Stochastic Depth''' works by randomly dropping paths in the residual blocks. On the <math>l^{th}</math> residual block the Stochastic Depth process is given as <math>G(x)=x+b_lF(x)</math> where <math>b_l \in \{0,1\}</math> is a Bernoulli random variable with probability <math>p_l</math>. Unlike sequential networks, there are many paths from the input to the output in these networks. By dropping some of the connections, the network is forced to flow through different paths to get the final deep layer representation. In a way it is similar to dropout, but for paths in multi-path networks. Using a constant value for <math>p_l</math> didn't work well, so instead a linear decay rule <math>p_l = 1 - \frac{l}{L}(1-p_L)</math> was used. In this equation, <math>L</math> is the number of layers, and <math>p_L</math> is the initial parameter. Essentially, the probability of a connection dropping in inversely proportional to the its depth in the network.<br />
<br />
'''Shake-Shake''' is a regularization method that specifically improves the ResNeXt (multiple residual connections) architecture. It is given as <math>G(x)=x+\alpha F_1(x)+(1-\alpha)F_2(x)</math>, where <math>\alpha \in [0,1]</math> is a random coefficient. Essentially, one of the parallel residual connections is dropped in the forward direction. This is similar to stochastic depth regularization, but a residual path always exists.<br />
Moreover, on the backward pass a similar random variable <math>\beta</math> is used to independently drop paths for gradient flow. This has the effect of adding noise in the gradients update process and improved performance over the vanilla ResNeXt network.<br />
<br />
<br />
[[File:Paper 32.jpg|600px|centre|thumb| Shake-Shake (ResNeXt + Shake-Shake) (Gastaldi, 2017), in which some processing layers omitted for conciseness.]]<br />
<br />
=Proposed Method=<br />
We give an intuitive interpretation of the forward pass of Shake-Shake regularization. To the best of our knowledge, it has not been given yet, while the phenomenon in the backward pass is experimentally investigated by Gastaldi (2017). In the forward pass, Shake-Shake interpolates the outputs of two residual branches with a random variable α that controls the degree of interpolation. As DeVries & Taylor (2017a) demonstrated that interpolation of two data in the feature space can synthesize reasonable augmented data, the interpolation of two residual blocks of Shake-Shake in the forward pass can be interpreted as synthesizing data. Use of a random variable α generates many different augmented data. On the other hand, in the backward pass, a different random variable β is used to disturb learning to make the network learnable long time. Gastaldi (2017) demonstrated how the difference between <math>\alpha</math> and <math>\beta</math> affects.<br />
<br />
The regularization mechanism of Shake-Shake relies on two or more residual branches, so that it can be applied only to 2-branch networks architectures. In addition, 2-branch network architectures consume more memory than 1-branch network architectures. One may think the number of learnable parameters of ResNeXt can be kept in 1-branch and 2-branch network architectures by controlling its cardinality and the number of channels (filters). For example, a 1-branch network (e.g., ResNeXt 1-64d) and its corresponding 2-branch network (e.g., ResNeXt 2-40d) have almost same number of learnable parameters. However, even so, it increases memory consumption due to the overhead to keep the inputs of residual blocks and so on. By comparing ResNeXt 1-64d and 2-40d, the latter requires more memory than the former by 8% in theory (for one layer) and by 11% in measured values (for 152 layers).<br />
<br />
This paper seeks to generalize the method proposed in Shake-Shake to be applied to any residual structure network. Shake-Shake. The initial formulation of 1-branch shake is <math>G(x) = x + \alpha F(x)</math>. In this case, <math>\alpha</math> is a coefficient that disturbs the forward pass, but is not necessarily constrained to be [0,1]. Another corresponding coefficient <math>\beta</math> is used in the backwards pass. Applying this simple adaptation of Shake-Shake on a 110-layer version of PyramidNet with <math>\alpha \in [0,1]</math> and <math>\beta \in [0,1]</math> performs abysmally, with an error rate of 77.99%.<br />
<br />
This failure is a result of the setup causing too much perturbation. A trick is needed to promote learning with large perturbations, to preserve the regularization effect. The idea of the authors is to borrow from ResDrop and combine that with Shake-Shake. This works by randomly deciding whether to apply 1-branch shake. This creates in effect two networks, the original network without a regularization component, and a regularized network. When mixing up two networks, we expected the following effects: When the non regularized network is selected, learning is promoted; when the perturbed network is selected, learning is disturbed. Achieving good performance requires a balance between the two. <br />
<br />
'''ShakeDrop''' is given as <br />
<br />
<div align="center"><br />
<math>G(x) = x + (b_l + \alpha - b_l \alpha)F(x)</math>,<br />
</div><br />
<br />
where <math>b_l</math> is a Bernoulli random variable following the linear decay rule used in Stochastic Depth. An alternative presentation is <br />
<br />
<div align="center"><br />
<math><br />
G(x) = \begin{cases}<br />
x + F(x) ~~ \text{if } b_l = 1 \\<br />
x + \alpha F(x) ~~ \text{otherwise}<br />
\end{cases}<br />
</math><br />
</div><br />
<br />
If <math>b_l = 1</math> then ShakeDrop is equivalent to the original network, otherwise it is the network + 1-branch Shake. The authors also found that the linear decay rule of ResDrop works well, compared with the uniform rule. Regardless of the value of <math>\beta</math> on the backwards pass, network weights will be updated.<br />
<br />
=Experiments=<br />
<br />
'''Parameter Search'''<br />
<br />
The authors experiments began with a hyperparameter search utilizing ShakeDrop on pyramidal networks. The PyramidNet used was made up of a total of 110 layers which included a convolutional layer and a final fully connected layer. It had 54 additive pyramidal residual blocks and the final residual block had 286 channels. The results of this search are presented below. <br />
<br />
[[File:ShakeDropHyperParameterSearch.png|600px|centre|thumb|Average Top-1 errors (%) of “PyramidNet + ShakeDrop” with several ranges of parameters of 4 runs at the final (300th) epoch on CIFAR-100 dataset in the “Batch” level. In some settings, it is equivalent to PyramidNet and PyramidDrop. Borrowed from ShakeDrop Regularization by Yamada et al., 2018.]]<br />
<br />
The setting that are used throughout the rest of the experiments are then <math>\alpha \in [-1,1]</math> and <math>\beta \in [0,1]</math>. Cases H and F outperform PyramidNet, suggesting that the strong perturbations imposed by ShakeDrop are functioning as intended. However, fully applying the perturbations in the backwards pass appears to destabilize the network, resulting in performance that is worse than standard PyramidNet.<br />
<br />
[[File:ParameterUpdateShakeDrop.png|400px|centre]]<br />
<br />
Following this initial parameter decision, the authors tested 4 different strategies for parameter update among "Batch" (same coefficients for all images in minibatch for each residual block), "Image" (same scaling coefficients for each image for each residual block), "Channel" (same scaling coefficients for each element for each residual block), and "Pixel" (same scaling coefficients for each element for each residual block). While Pixel was the best in terms of error rate, it is not very memory efficient, so Image was selected as it had the second best performance without the memory drawback.<br />
<br />
'''Comparison with Regularization Methods'''<br />
<br />
For these experiments, there are a few modifications that were made to assist with training. For ResNeXt, the EraseRelu formulation has each residual block ends in batch normalization. The Wide ResNet also is compared between vanilla with batch normalization and without. Batch normalization keeps the outputs of residual blocks in a certain range, as otherwise <math>\alpha</math> and <math>\beta</math> could cause perturbations that are too large, causing divergent learning. There is also a comparison of ResDrop/ShakeDrop Type A (where the regularization unit is inserted before the add unit for a residual branch) and after (where the regularization unit is inserted after the add unit for a residual branch). <br />
<br />
These experiments are performed on the CIFAR-100 dataset.<br />
<br />
[[File:ShakeDropArchitectureComparison1.png|800px|centre|thumb|]]<br />
<br />
[[File:ShakeDropArchitectureComparison2.png|800px|centre|thumb|]]<br />
<br />
[[File:ShakeDropArchitectureComparison3.png|800px|centre|thumb|]]<br />
<br />
For a final round of testing, the training setup was modified to incorporate other techniques used in state of the art methods. For most of the tests, the learning rate for the 300 epoch version started at 0.1 and decayed by a factor of 0.1 1/2 & 3/4 of the way through training. The alternative was cosine annealing, based on the presentation by Loshchilov and Hutter in their paper SGDR: Stochastic Gradient Descent with Warm Restarts. This is indicated in the Cos column, with a check indicating cosine annealing. <br />
<br />
[[File:CosineAnnealing.png|400px|centre|thumb|]]<br />
<br />
The Reg column indicates the regularization method used, either none, ResDrop (RD), Shake-Shake (SS), or ShakeDrop (SD). Fianlly, the Fil Column determines the type of data augmentation used, either none, cutout (CO) (DeVries & Taylor, 2017b), or Random Erasing (RE) (Zhong et al., 2017). <br />
<br />
[[File:ShakeDropComparison.png|800px|centre|thumb|Top-1 Errors (%) at final epoch on CIFAR-10/100 datasets]]<br />
<br />
'''State-of-the-Art Comparisons'''<br />
<br />
A direct comparison with state of the art methods is favorable for this new method. <br />
<br />
# Fair comparison of ResNeXt + Shake-Shake with PyramidNet + ShakeDrop gives an improvement of 0.19% on CIFAR-10 and 1.86% on CIFAR-100. Under these conditions, the final error rate is then 2.67% for CIFAR-10 and 13.99% for CIFAR-100.<br />
# Fair comparison of ResNeXt + Shake-Shake + Cutout with PyramidNet + ShakeDrop + Random Erasing gives an improvement of 0.25% on CIFAR-10 and 3.01% on CIFAR 100. Under these conditions, the final error rate is then 2.31% for CIFAR-10 and 12.19% for CIFAR-100.<br />
# Comparison with the state-of-the-arts, PyramidNet + ShakeDrop gives an improvement of 0.25% on CIFAR-10 than ResNeXt + Shake-Shake + Cutout, PyramidNet + ShakeDrop gives an improvement of 2.85% on CIFAR-100 than Coupled Ensemble.<br />
<br />
=Implementation details=<br />
<br />
'''CIFAR-10/100 datasets'''<br />
<br />
All the images in these datasets were color normalized and then horizontally flipped with a probability of 50%. All of then then were zero padded to have a dimentionality of 40 by 40 pixels.<br />
<br />
<br />
=Conclusion=<br />
The paper proposes a new form of regularization that is an extension of "Shake-Shake" regularization [Gastaldi, 2017]. The original "shake-shake" proposes using two residual paths adding to the same output, and during training, considering different randomly selected convex combinations of the two paths (while using an equally weighted combination at test time). This paper contends that this requires additional memory, and attempts to achieve similar regularization with a single path. To do so, they train a network with a single residual path, where the residual is included without attenuation in some cases with some fixed probability, and attenuated randomly (or even inverted) in others. The paper contends that this achieves superior performance than choosing simply a random attenuation for every sample (although, this can be seen as choosing an attenuation under a distribution with some fixed probability mass.<br />
<br />
Their stochastic regularization method, ShakeDrop, which outperforms previous state of the art methods while maintaining similar memory efficiency. It demonstrates that heavily perturbing a network can help to overcome issues with overfitting. It is also an effective way to regularize residual networks for image classification. The method was tested by CIFAR-10/100 and Tiny ImageNet datasets and showed great performance.<br />
<br />
=Critique=<br />
<br />
The novelty of this paper is low as pointed out by the reviewers. Also, there is a confusion whether or not the results could be replicated as <math>\alpha</math> and <math>\beta</math> are choosen randomly. The proposed ShakeDrop regularization is essentially a combination of the PyramidDrop and Shake-Shake regularization. The most surprising part is that the forward weight can be negative thus inverting the output of a convolution. The mathematical justification for ShakeDrop regularization is limited, relying on intuition and empirical evidence instead.<br />
<br />
One downside of this methods (as was identified in the presentation as well) is that the training for cosine annealing variation of the model takes 1800 epochs which is time intensive compared to other methods that were compared as baselines. This can limit practical implementation of this algorithm.<br />
<br />
As pointed out from the above, the method basically relies heavily on the intuition. This means that the performance of the algorithm can not been extended beyond the CIFAR dataset and can vary a lot depending on the characteristics of data sets that users are performing, with some exaggeration. However, the performance is still impressive since it performs better than known algorithms. It is not clear as to how the proposed technique would work with a non-residual architecture.<br />
It lacks conclusive proof that "shake-drop" is a generically useful regularization technique. For one, the method is evaluated only on small toy-datasets: CIFAR-10 and CIFAR-100. Evaluation on Imagenet perhaps would have been valuable. There is also another dataset that would of been good to try SVHN. Overall I believe the impact of this beyond CIFAR is unclear.<br />
<br />
=References=<br />
[Yamada et al., 2018] Yamada Y, Iwamura M, Kise K. ShakeDrop regularization. arXiv preprint arXiv:1802.02375. 2018 Feb 7.<br />
<br />
[He et al., 2016] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proc. CVPR, 2016.<br />
<br />
[Zagoruyko & Komodakis, 2016] Sergey Zagoruyko and Nikos Komodakis. Wide residual networks. In Proc. BMVC, 2016.<br />
<br />
[Han et al., 2017] Dongyoon Han, Jiwhan Kim, and Junmo Kim. Deep pyramidal residual networks. In Proc. CVPR, 2017a.<br />
<br />
[Xie et al., 2017] Saining Xie, Ross Girshick, Piotr Dollar, Zhuowen Tu, and Kaiming He. Aggregated residual transformations for deep neural networks. In Proc. CVPR, 2017.<br />
<br />
[Huang et al., 2016] Gao Huang, Yu Sun, Zhuang Liu, Daniel Sedra, and Kilian Weinberger. Deep networks with stochastic depth. arXiv preprint arXiv:1603.09382v3, 2016.<br />
<br />
[Gastaldi, 2017] Xavier Gastaldi. Shake-shake regularization. arXiv preprint arXiv:1705.07485v2, 2017.<br />
<br />
[Loshilov & Hutter, 2016] Ilya Loshchilov and Frank Hutter. Sgdr: Stochastic gradient descent with warm restarts. arXiv preprint arXiv:1608.03983, 2016.<br />
<br />
[DeVries & Taylor, 2017b] Terrance DeVries and Graham W. Taylor. Improved regularization of convolutional neural networks with cutout. arXiv preprint arXiv:1708.04552, 2017b.<br />
<br />
[Zhong et al., 2017] Zhun Zhong, Liang Zheng, Guoliang Kang, Shaozi Li, and Yi Yang. Random erasing data augmentation. arXiv preprint arXiv:1708.04896, 2017.<br />
<br />
[Dutt et al., 2017] Anuvabh Dutt, Denis Pellerin, and Georges Qunot. Coupled ensembles of neural networks. arXiv preprint 1709.06053v1, 2017.<br />
<br />
[Veit et al., 2016] Andreas Veit, Michael J Wilber, and Serge Belongie. Residual networks behave like ensembles of relatively shallow networks. Advances in Neural Information Processing Systems 29, 2016.</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=MULTI-VIEW_DATA_GENERATION_WITHOUT_VIEW_SUPERVISION&diff=42348MULTI-VIEW DATA GENERATION WITHOUT VIEW SUPERVISION2018-12-08T06:32:23Z<p>Gchalato: /* Evaluation of the Quality of Generated Samples */</p>
<hr />
<div>This page contains a summary of the paper "[https://openreview.net/forum?id=ryRh0bb0Z Multi-View Data Generation without Supervision]" by Mickael Chen, Ludovic Denoyer, Thierry Artieres. It was published at the International Conference on Learning Representations (ICLR) in 2018. An implementation of the models presented in this paper is available here[https://github.com/mickaelChen/GMV]<br />
<br />
==Introduction==<br />
<br />
===Motivation===<br />
High Dimensional Generative models have seen a surge of interest of late with the introduction of Variational Auto-Encoders and Generative Adversarial Networks. This paper focuses on a particular problem where one aims at generating samples corresponding to a number of objects under various views. The distribution of the data is assumed to be driven by two independent latent factors: the content, which represents the intrinsic features of an object, and the view, which stands for the settings of a particular observation of that object (for example, the different angles of the same object). The paper proposes two models using this disentanglement of latent space - a generative model and a conditional variant of the same. The authors claim that unlike many multiview<br />
approaches, the proposed model doesn’t need any supervision on the views but only on the content.<br />
<br />
===Related Work===<br />
<br />
The problem of handling multi-view inputs has mainly been studied from the predictive point of view where one wants, for example, to learn a model able to predict/classify over multiple views of the same object (Su et al. (2015); Qi et al. (2016)). These approaches generally involve (early or late) fusion of the different views at a particular level of a deep architecture. Recent studies have focused on identifying factors of variations from multiview datasets. The underlying idea is to consider that a particular data sample may be thought as the mix of a content information (e.g. related to its class label like a given person in a face dataset) and of a side information, the view, which accounts for factors of variability (e.g. exposure, viewpoint, with/wo glasses...). So, all the samples of the same class contain the same content but different view. A number of approaches have been proposed to disentangle the content from the view (i.e. methods based on unlabeled samples), also referred as the style in some papers (Mathieu et al. (2016); Denton & Birodkar (2017)). The two common limitations the earlier approaches pose - as claimed by the paper - are that (i) they usually<br />
consider discrete views that are characterized by a domain or a set of discrete (binary/categorical) attributes (e.g. face with/wo glasses, the color of the hair, etc.) and could not easily scale to a large number of attributes or to continuous views. (ii) most models are trained using view supervision (e.g. the view attributes), which of course greatly helps in the learning of such model, yet prevents their use on many datasets where this information is not available.<br />
<br />
Recently such attempts have been made to learn such models without supervision, but they cannot disentangle high level concepts as only simple features can be reliably captured without any guidance.<br />
<br />
===Contributions===<br />
<br />
The contributions that authors claim are the following: (i) A new generative model able to generate data with various content and high view diversity using a supervision on the content information only. (ii) Extend the generative model to a conditional model that allows generating new views over any input sample. (iii) Report experimental results on four different images datasets to prove that the models can generate realistic samples and capture (and generate with) the diversity of views.<br />
<br />
Precisely,two models have been proposed:<br />
# a generative model ('''GMV - Generative Multi-view Model''') that generates objects under various views (multiview generation), <br />
# and a conditional extension, '''conditional GMV (C-GMV)''' of this model that generates a large number of views of any input object (conditional multi-view generation). <br />
<br />
Both models are based on the adversarial training schema of Generative Adversarial Networks (GAN) proposed in Goodfellow et al. (2014)). The simple but strong idea is to focus on distributions over pairs of examples (e.g. images representing a same object in different views) rather than distribution on single examples.<br />
<br />
==Paper Overview==<br />
<br />
===Background===<br />
<br />
The paper uses the concept of the popular GAN (Generative Adverserial Networks) proposed by Goodfellow et al.(2014).<br />
<br />
GENERATIVE ADVERSARIAL NETWORK:<br />
<br />
Generative adversarial networks (GANs) are deep neural net architectures comprised of two nets, pitting one against the other (thus the “adversarial”). GANs was introduced in a paper by Ian Goodfellow and other researchers at the University of Montreal, including Yoshua Bengio, in 2014. Referring to GANs, Facebook’s AI research director Yann LeCun called adversarial training “the most interesting idea in the last 10 years in ML.”<br />
<br />
Let us denote <math>X</math> an input space composed of multidimensional samples <math>x</math> e.g. vector, matrix or tensor. Given a latent space <math>R^n</math> and a prior distribution <math>p_z(z)</math> over this latent space, any generator function <math>G : R^n → X</math> defines a distribution <math>p_G </math> on <math> X</math> which is the distribution of samples <math>G(z)</math> where <math>z ∼ p_z</math>. A GAN defines, in addition to <math>G</math>, a discriminator function <math>D : X → [0; 1]</math> which aims at differentiating between real inputs sampled from the training set and fake inputs sampled from <math>p_G</math>, while the generator learns to fool the discriminator <math>D</math>. Usually both <math>G</math> and <math>D</math> are implemented with neural networks. The objective function is based on the following adversarial criterion:<br />
<br />
<div style="text-align: center;font-size:100%"><math>\underset{G}{min} \ \underset{D}{max}</math> <math>E_{p_x}[log D(x)] + Ep_z[log(1 − D(G(z)))]</math></div><br />
<br />
where <math>p_x</math> is the empirical data distribution on <math>X</math> .<br />
It has been shown in Goodfellow et al. (2014) that if G∗ and D∗ are optimal for the above criterion, the Jensen-Shannon divergence between <math>p_{G∗}</math> and the empirical distribution of the data <math>p_x</math> in the dataset is minimized, making GAN able to estimate complex continuous data distributions.<br />
<br />
CONDITIONAL GENERATIVE ADVERSARIAL NETWORK:<br />
<br />
In the Conditional GAN (CGAN), the generator learns to generate a fake sample with a specific condition or characteristics (such as a label associated with an image or more detailed tag) rather than a generic sample from unknown noise distribution. The conditionality of a CGAN is determined by defining a generator function <math>G</math> which takes a noise vector <math>z</math> and a condition <math>y</math> as inputs. Now, to add such a condition to both generator and discriminator, we will simply feed some vector <math>y</math>, into both networks. Hence, both the discriminator <math>D(X,y)</math> and generator <math>G(z,y)</math> are jointly distributed with <math>y</math>. A target <math>X</math> from a given input <math>y</math> can be obtained by first sampling the latent vector <math>z ∼ p_z</math>, then by computing <math>G(y, z)</math>. The discriminator takes both the condition <math>y</math> and the datapoint <math>x</math> as inputs.<br />
<br />
Now, the objective function of CGAN is:<br />
<br />
<div style="text-align: center;font-size:100%"><math>\underset{G}{min} \ \underset{D}{max}</math> <math>E_{p_x}[log D(x,y)] + Ep_z[log(1 − D(G(y,z)))]</math></div><br />
<br />
The paper also suggests that many studies have reported that when dealing with high-dimensional input spaces, CGAN tends to collapse the modes of the data distribution, mostly ignoring the latent factor <math>z</math> and generating <math>x</math> only based on the condition <math>y</math>, exhibiting an almost deterministic behavior. At this point, the CGAN also fails to produce a satisfying amount of diversity in generated samples.<br />
<br />
===Generative Multi-View Model===<br />
<br />
''' Objective and Notations: ''' The distribution of the data x ∈ X is assumed to be driven by two latent factors: a content factor denoted c which corresponds to the invariant proprieties of the object and a view factor denoted v which corresponds to the factor of variations. Typically, if X is the space of people’s faces, c stands for the intrinsic features of a person’s face while v stands for the transient features and the viewpoint of a particular photo of the face, including the photo exposure<br />
and additional elements like a hat, glasses, etc.... These two factors c and v are assumed to be independent and these are the factors needed to learn.<br />
<br />
The paper defines two tasks here to be done: <br />
(i) '''Multi View Generation''': we want to be able to sample over X by controlling the two factors c and v. Given two priors, p(c) and p(v), this sampling will be possible if we are able to estimate p(x|c, v) from a training set.<br />
(ii) '''Conditional Multi-View Generation''': the second objective is to be able to sample different views of a given object. Given a prior p(v), this sampling will be achieved by learning the probability p(c|x), in addition to p(x|c, v). Ability to learn generative models able to generate from a disentangled latent space would allow controlling the sampling on the two different axes,<br />
the content and the view. The authors claim the originality of work is to learn such generative models without using any view labeling information.<br />
<br />
The paper introduces the vectors '''c''' and '''v''' to represent latent vectors in R<sup>c</sup> and R<sup>v</sup><br />
<br />
<br />
''' Generative Multi-view Model: '''<br />
<br />
Consider two prior distributions over the content and view factors denoted as <math>p_c</math> and <math>p_v</math>, corresponding to the prior distribution over content and latent factors. Moreover, we consider a generator G that implements a distribution over samples x, denoted as <math>p_G</math> by computing G(c, v) with <math>c ∼ p_c</math> and <math>v ∼ p_v</math>. The objective is to learn this generator so that its first input c corresponds to the content of the generated sample while its second input v, captures the underlying view of the sample. Doing so would allow one to control the output sample of the generator by tuning its content or its view (i.e. c and v).<br />
<br />
The key idea that authors propose is to focus on the distribution of pairs of inputs rather than on the distribution over individual samples. When no view supervision is available the only valuable pairs of samples that one may build from the dataset consist of two samples of a given object under two different views. When we choose any two samples randomly from the dataset from the same object, it is most likely that we get two different views. The paper explains that there are three goals here, (i) As in regular GAN, each sample generated by G needs to look realistic. (ii) As real pairs are composed of two views of the same object, the generator should generate pairs of the same object. Since the two sampled view factors v1 and v2 are different, the only way this can be achieved is by encoding the content vector c which is invariant. (iii) It is expected that the discriminator should easily discriminate between a pair of samples corresponding to the same object under different views from a pair of samples corresponding to a same object under the same view. Because the pair shares the same content factor c, this should force the generator to use the view factors v1 and v2 to produce diversity in the generated pair.<br />
<br />
Now, the objective function of GMV Model is:<br />
<br />
<div style="text-align: center;font-size:100%"><math>\underset{G}{min} \ \underset{D}{max}</math> <math>E_{x_1,x_2}[log D(x_1,x_2)] + E_{v_1,v_2}[log(1 − D(G(c,v_1),G(c,v_2)))]</math></div><br />
<br />
Once the model is learned, generator G that generates single samples by first sampling c and v following <math>p_c</math> and <math>p_v</math>, then by computing G(c, v). By freezing c or v, one may then generate samples corresponding to multiple views of any particular content, or corresponding to many contents under a particular view. One can also make interpolations between two given views over a particular content, or between two contents using a particular view<br />
<br />
<div style="text-align: center;font-size:100%">[[File:GMV.png]]</div><br />
<br />
===Conditional Generative Model (C-GMV)===<br />
<br />
C-GMV is proposed by the authors to be able to change the view of a given object that would be provided as an input to the model. This model extends the generative model's the ability to extract the content factor from any given input and to use this extracted content in order to generate new views of the corresponding object. To achieve such a goal, we must add to our generative model an encoder function denoted <math>E : X → R^C</math> that will map any input in X to the content space <math>R^C</math><br />
<br />
Input sample x is encoded in the content space using an encoder function, noted E (implemented as a neural network).<br />
This encoder serves to generate a content vector c = E(x) that will be combined with a randomly sampled view <math>v ∼ p_v</math> to generate an artificial example. The artificial sample is then combined with the original input x to form a negative pair. The issue with this approach is that CGAN is known to easily miss modes of the underlying distribution. The generator enters in a state where it ignores the noisy component v. To overcome this phenomenon, we use the same idea as in GMV. We build negative pairs <math>(G(c, v_1), G(c, v_2))</math> by randomly sampling two views <math>v_1</math> and <math>v_2</math> that are combined to get a unique content c. c is computed from a sample x using the encoder E, i.e. c= E(x). By doing so, the ability of our approach to generating pairs with view diversity is preserved. Since this diversity can only be captured by taking into account the two different view vectors provided to the model (<math>v_1</math> and <math>v_2</math>), this will encourage G(c, v) to generate samples containing both the content information c, and the view v. Positive pairs are sampled from the training set and correspond to two views of a given object.<br />
<br />
The Objective function for C-GMV will be:<br />
<br />
<div style="text-align: center;font-size:100%"><math>\underset{G}{min} \ \underset{D}{max}</math> <math>E_{x_1,x_2 ~ p_x|l(x_1)=l(x_2)}[log D(x_1,x_2)] + E_{v_1,v_2 ~ p_v,x~p_x}[log(1 − D(G(E(x),v_1),G(E(x),v_2)))]+E_{v∼p_v,x∼p_x}[log(1 − D(G(E(x), v), x))] </math></div><br />
<br />
<div style="text-align: center;font-size:100%">[[File:CGMV.png]]</div><br />
<br />
<br />
At inference time, as with the GMV model, we are interested in getting the encoder E and the<br />
generator G. These models may be used for generating new views of any object which is observed<br />
as an input sample x by computing its content vector E(x), then sampling <math>v ∼ p_v</math> and finally by<br />
computing the output G(E(x), v)<br />
<br />
==Experiments and Results==<br />
<br />
The authors have given an exhaustive set of results and experiments.<br />
<br />
Datasets: The two models were evaluated by performing experiments over four image datasets of various domains. Note that when supervision is available on the views (like CelebA for example where images are labeled with attributes) it is not used for learning models. The only supervision that is used is if two samples correspond to the same object or not.<br />
<br />
<div style="text-align: center;font-size:100%">[[File:table_data.png]]</div><br />
<br />
<br />
Model Architecture: Same architectures for every dataset. The images were rescaled to 3×64×64 tensors. The generator G and the discriminator D follow that of the DCGAN implementation proposed in Radford et al. (2015). The encoder E is similar to D with the only differences being the batch-normalization in the first layer and the last layer which doesn't have a non-linearity. The Adam optimizer was used, with a batch size of 128. The learning rates for G and D were set to 1*10<sup>-3</sup> and 2*10<sup>-4</sup> respectively for the GMV experiments. In the C-GMV experiments, learning rates of 5*10<sup>-5</sup> were used. Alternating gradient descent was used to optimize the different objectives of the network components (generator, encoder and discriminator).<br />
<br />
Baselines: Most existing methods are learned on datasets with view labeling. To fairly compare with alternative models, authors have built baselines working in the same conditions as the models in this paper. In addition, models are compared with the model from Mathieu et al. (2016). Results gained with two implementations are reported, the first one based on the implementation provided by the authors2 (denoted Mathieu et al. (2016)), and the second one (denoted Mathieu et al. (2016) (DCGAN) ) that implements the same model using architectures inspired from DCGAN Radford et al. (2015), which is more stable and that was tuned to allow a fair comparison with our approach. For pure multi-view generative setting, generative model(GMV) is compared with standard GANs that are learned to approximate the joint generation of multiple samples: DCGANx2 is learned to output pairs of views over the same object, DCGANx4 is trained on quadruplets, and DCGANx8 on eight different views. <br />
<br />
===Generating Multiple Contents and Views===<br />
<br />
Figure 1 shows examples of generated images by our model and Figure 4 shows images sampled by the DCGAN based models (DCGANx2, DCGANx4, and DCGANx8) on 3DChairs and CelebA datasets.<br />
<br />
<div style="text-align: center;font-size:100%">[[File:fig1_gmv.png]]</div><br />
<br />
<div style="text-align: center;font-size:100%">[[File:fig4_gmv.png]]</div><br />
<br />
<br />
Figure 5 shows additional results, using the same presentation, for the GMV model only on two other datasets. In the left hand block of Figure 5, each row shows different views generated given the same content. <br />
<br />
<div style="text-align: center;font-size:100%">[[File:fig5_gmv.png]]</div><br />
<br />
Figure 6 shows generated samples obtained by interpolation between two different view factors (left) or two content factors (right). Again, in the left and right hand block of Figure 6, each row shows different views generated given the same content. It allows us to have a better idea of the underlying view/content structure captured by GMV. We can see that our approach is able to smoothly move from one content/view to another content/view while keeping the other factor constant. This also illustrates that content and view factors are well independently handled by the generator i.e. changing the view<br />
does not modify the content and vice versa.<br />
<br />
<br />
<div style="text-align: center;font-size:100%">[[File:fig6_gmv.png]]</div><br />
<br />
===Generating Multiple Views of a Given Object===<br />
<br />
The second set of experiments evaluates the ability of C-GMV to capture a particular content from an input sample and to use this content to generate multiple views of the same object. Figure 7 and 8 illustrate the diversity of views in samples generated by our model and compare our results with those obtained with the CGAN model and to models from Mathieu et al. (2016). For each row, the input sample is shown in the left column. New views are generated from that input and shown to the right, with those generated from C_GMV in the centre, and those generated from CGAN on the far right.<br />
<br />
<div style="text-align: center;font-size:100%">[[File:fig7_gmv.png]]</div><br />
<br />
<br />
<div style="text-align: center;font-size:100%">[[File:fig8_gmv.png]]</div><br />
<br />
=== Evaluation of the Quality of Generated Samples ===<br />
<br />
There are usually several metrics to evaluate generative models. Some of them are: <br />
<ol><br />
<li>Inception Score: In a general sense, the Inception Score is a metric used to quantify the “realness” of a generated image. It is calculated across a set of generated images, and considers two criteria. First, all images of the sample class should be similar (low in-class variance). And second, the distribution of classes should not be dominated by any particular class. The better these criteria are met; the higher the Inception Score.</li><br />
<li>Latent Space Interpolation</li><br />
<li>log-likelihood (LL) score</li><br />
<li> minimum description length (MDL) score</li><br />
<li>minimum message length (MML) score</li><br />
<li>Akaike Information Criterion (AIC) score</li><br />
<li>Bayesian Information Criterion (BIC) score</li><br />
</ol><br />
<br />
<br />
<br />
<br />
<br />
The authors did sets of experiments aimed at evaluating the quality of the generated samples. They have been made on the CelebA dataset and evaluate (i) the ability of the models to preserve the identity of a person in multiple generated views, (ii) to generate realistic samples, (iii) to preserve the diversity in the generated views and (iv) to capture the view distributions of the original dataset.<br />
<br />
<div style="text-align: center;font-size:100%">[[File:tab3.png]]</div><br />
<br />
<br />
<div style="text-align: center;font-size:100%">[[File:tab4.png]]</div><br />
<br />
<br />
<div style="text-align: center;font-size:100%">[[File:table.png]]</div><br />
<br />
==Conclusion==<br />
<br />
The paper proposed a generative model, which can be learnt from multi-view data without any supervision. Moreover, it introduced a conditional version that allows generating new views of an input image. Using experiments, they proved that the model can capture content and view factors. Here, the paper showed that the application of architecture search to dense image prediction was achieved through a) The construction of a recursive search space leveraging innovation in the dense prediction literature b) construction of a fast proxy predictive of a large task. The learned architecture was shown to surpass human invented architectures across three dense image prediction tasks i.e scene parsing, person part segmentation and semantic segmentation. In the future, they are planning to use the method of this paper for data augmentation which can enrich training dataset. .<br />
<br />
==Future Work==<br />
The authors of the papers mentioned that they plan to explore using their model for data augmentation, as it can produce other data views for training, in both semi-supervised and one-shot/few-shot learning settings. <br />
<br />
==Critique==<br />
<br />
The main idea is to train the model with pairs of images with different views. It is not that clear as to what defines a view in particular. The algorithms are largely based on earlier concepts of GAN and CGAN The authors give reference to the previous papers tackling the same problem and clearly define that the novelty in this approach is not making use of view labels. The authors give a very thorough list of experiments which clearly establish the superiority of the proposed models to baselines.<br />
<br />
However, this paper only tested the model on rather constrained examples. As was observed in the results the proposed approach seems to have a high sample complexity relying on training samples covering the full range of variations for both specified and unspecified variations. Also, the proposed model does not attempt to disentangle variations within the specified and unspecified components.<br />
<br />
The method that the paper presented is novel and the paper is easy to follow. However, the authors only show a comparison between the proposed method and several baselines: DCGAN and CGAN and do not compare with the methods from Mathieu et al. 2016. In addition, the experiment result is empirical, we do not know the performance of this method in practice in the real word.<br />
<br />
==References==<br />
<br />
[1] Mickael Chen, Ludovic Denoyer, Thierry Artieres. MULTI-VIEW DATA GENERATION WITHOUT VIEW SUPERVISION. Published as a conference paper at ICLR 2018<br />
<br />
[2] Michael F Mathieu, Junbo Jake Zhao, Junbo Zhao, Aditya Ramesh, Pablo Sprechmann, and Yann LeCun. Disentangling factors of variation in deep representation using adversarial training. In Advances in Neural Information Processing Systems, pp. 5040–5048, 2016.<br />
<br />
[3] Mathieu Aubry, Daniel Maturana, Alexei Efros, Bryan Russell, and Josef Sivic. Seeing 3d chairs: exemplar part-based 2d-3d alignment using a large dataset of cad models. In CVPR, 2014.<br />
<br />
[4] Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Advances in neural information processing systems, pp. 2672–2680, 2014.<br />
<br />
[5] Emily Denton and Vighnesh Birodkar. Unsupervised learning of disentangled representations from video. arXiv preprint arXiv:1705.10915, 2017.</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=MULTI-VIEW_DATA_GENERATION_WITHOUT_VIEW_SUPERVISION&diff=42347MULTI-VIEW DATA GENERATION WITHOUT VIEW SUPERVISION2018-12-08T06:32:00Z<p>Gchalato: /* Evaluation of the Quality of Generated Samples */</p>
<hr />
<div>This page contains a summary of the paper "[https://openreview.net/forum?id=ryRh0bb0Z Multi-View Data Generation without Supervision]" by Mickael Chen, Ludovic Denoyer, Thierry Artieres. It was published at the International Conference on Learning Representations (ICLR) in 2018. An implementation of the models presented in this paper is available here[https://github.com/mickaelChen/GMV]<br />
<br />
==Introduction==<br />
<br />
===Motivation===<br />
High Dimensional Generative models have seen a surge of interest of late with the introduction of Variational Auto-Encoders and Generative Adversarial Networks. This paper focuses on a particular problem where one aims at generating samples corresponding to a number of objects under various views. The distribution of the data is assumed to be driven by two independent latent factors: the content, which represents the intrinsic features of an object, and the view, which stands for the settings of a particular observation of that object (for example, the different angles of the same object). The paper proposes two models using this disentanglement of latent space - a generative model and a conditional variant of the same. The authors claim that unlike many multiview<br />
approaches, the proposed model doesn’t need any supervision on the views but only on the content.<br />
<br />
===Related Work===<br />
<br />
The problem of handling multi-view inputs has mainly been studied from the predictive point of view where one wants, for example, to learn a model able to predict/classify over multiple views of the same object (Su et al. (2015); Qi et al. (2016)). These approaches generally involve (early or late) fusion of the different views at a particular level of a deep architecture. Recent studies have focused on identifying factors of variations from multiview datasets. The underlying idea is to consider that a particular data sample may be thought as the mix of a content information (e.g. related to its class label like a given person in a face dataset) and of a side information, the view, which accounts for factors of variability (e.g. exposure, viewpoint, with/wo glasses...). So, all the samples of the same class contain the same content but different view. A number of approaches have been proposed to disentangle the content from the view (i.e. methods based on unlabeled samples), also referred as the style in some papers (Mathieu et al. (2016); Denton & Birodkar (2017)). The two common limitations the earlier approaches pose - as claimed by the paper - are that (i) they usually<br />
consider discrete views that are characterized by a domain or a set of discrete (binary/categorical) attributes (e.g. face with/wo glasses, the color of the hair, etc.) and could not easily scale to a large number of attributes or to continuous views. (ii) most models are trained using view supervision (e.g. the view attributes), which of course greatly helps in the learning of such model, yet prevents their use on many datasets where this information is not available.<br />
<br />
Recently such attempts have been made to learn such models without supervision, but they cannot disentangle high level concepts as only simple features can be reliably captured without any guidance.<br />
<br />
===Contributions===<br />
<br />
The contributions that authors claim are the following: (i) A new generative model able to generate data with various content and high view diversity using a supervision on the content information only. (ii) Extend the generative model to a conditional model that allows generating new views over any input sample. (iii) Report experimental results on four different images datasets to prove that the models can generate realistic samples and capture (and generate with) the diversity of views.<br />
<br />
Precisely,two models have been proposed:<br />
# a generative model ('''GMV - Generative Multi-view Model''') that generates objects under various views (multiview generation), <br />
# and a conditional extension, '''conditional GMV (C-GMV)''' of this model that generates a large number of views of any input object (conditional multi-view generation). <br />
<br />
Both models are based on the adversarial training schema of Generative Adversarial Networks (GAN) proposed in Goodfellow et al. (2014)). The simple but strong idea is to focus on distributions over pairs of examples (e.g. images representing a same object in different views) rather than distribution on single examples.<br />
<br />
==Paper Overview==<br />
<br />
===Background===<br />
<br />
The paper uses the concept of the popular GAN (Generative Adverserial Networks) proposed by Goodfellow et al.(2014).<br />
<br />
GENERATIVE ADVERSARIAL NETWORK:<br />
<br />
Generative adversarial networks (GANs) are deep neural net architectures comprised of two nets, pitting one against the other (thus the “adversarial”). GANs was introduced in a paper by Ian Goodfellow and other researchers at the University of Montreal, including Yoshua Bengio, in 2014. Referring to GANs, Facebook’s AI research director Yann LeCun called adversarial training “the most interesting idea in the last 10 years in ML.”<br />
<br />
Let us denote <math>X</math> an input space composed of multidimensional samples <math>x</math> e.g. vector, matrix or tensor. Given a latent space <math>R^n</math> and a prior distribution <math>p_z(z)</math> over this latent space, any generator function <math>G : R^n → X</math> defines a distribution <math>p_G </math> on <math> X</math> which is the distribution of samples <math>G(z)</math> where <math>z ∼ p_z</math>. A GAN defines, in addition to <math>G</math>, a discriminator function <math>D : X → [0; 1]</math> which aims at differentiating between real inputs sampled from the training set and fake inputs sampled from <math>p_G</math>, while the generator learns to fool the discriminator <math>D</math>. Usually both <math>G</math> and <math>D</math> are implemented with neural networks. The objective function is based on the following adversarial criterion:<br />
<br />
<div style="text-align: center;font-size:100%"><math>\underset{G}{min} \ \underset{D}{max}</math> <math>E_{p_x}[log D(x)] + Ep_z[log(1 − D(G(z)))]</math></div><br />
<br />
where <math>p_x</math> is the empirical data distribution on <math>X</math> .<br />
It has been shown in Goodfellow et al. (2014) that if G∗ and D∗ are optimal for the above criterion, the Jensen-Shannon divergence between <math>p_{G∗}</math> and the empirical distribution of the data <math>p_x</math> in the dataset is minimized, making GAN able to estimate complex continuous data distributions.<br />
<br />
CONDITIONAL GENERATIVE ADVERSARIAL NETWORK:<br />
<br />
In the Conditional GAN (CGAN), the generator learns to generate a fake sample with a specific condition or characteristics (such as a label associated with an image or more detailed tag) rather than a generic sample from unknown noise distribution. The conditionality of a CGAN is determined by defining a generator function <math>G</math> which takes a noise vector <math>z</math> and a condition <math>y</math> as inputs. Now, to add such a condition to both generator and discriminator, we will simply feed some vector <math>y</math>, into both networks. Hence, both the discriminator <math>D(X,y)</math> and generator <math>G(z,y)</math> are jointly distributed with <math>y</math>. A target <math>X</math> from a given input <math>y</math> can be obtained by first sampling the latent vector <math>z ∼ p_z</math>, then by computing <math>G(y, z)</math>. The discriminator takes both the condition <math>y</math> and the datapoint <math>x</math> as inputs.<br />
<br />
Now, the objective function of CGAN is:<br />
<br />
<div style="text-align: center;font-size:100%"><math>\underset{G}{min} \ \underset{D}{max}</math> <math>E_{p_x}[log D(x,y)] + Ep_z[log(1 − D(G(y,z)))]</math></div><br />
<br />
The paper also suggests that many studies have reported that when dealing with high-dimensional input spaces, CGAN tends to collapse the modes of the data distribution, mostly ignoring the latent factor <math>z</math> and generating <math>x</math> only based on the condition <math>y</math>, exhibiting an almost deterministic behavior. At this point, the CGAN also fails to produce a satisfying amount of diversity in generated samples.<br />
<br />
===Generative Multi-View Model===<br />
<br />
''' Objective and Notations: ''' The distribution of the data x ∈ X is assumed to be driven by two latent factors: a content factor denoted c which corresponds to the invariant proprieties of the object and a view factor denoted v which corresponds to the factor of variations. Typically, if X is the space of people’s faces, c stands for the intrinsic features of a person’s face while v stands for the transient features and the viewpoint of a particular photo of the face, including the photo exposure<br />
and additional elements like a hat, glasses, etc.... These two factors c and v are assumed to be independent and these are the factors needed to learn.<br />
<br />
The paper defines two tasks here to be done: <br />
(i) '''Multi View Generation''': we want to be able to sample over X by controlling the two factors c and v. Given two priors, p(c) and p(v), this sampling will be possible if we are able to estimate p(x|c, v) from a training set.<br />
(ii) '''Conditional Multi-View Generation''': the second objective is to be able to sample different views of a given object. Given a prior p(v), this sampling will be achieved by learning the probability p(c|x), in addition to p(x|c, v). Ability to learn generative models able to generate from a disentangled latent space would allow controlling the sampling on the two different axes,<br />
the content and the view. The authors claim the originality of work is to learn such generative models without using any view labeling information.<br />
<br />
The paper introduces the vectors '''c''' and '''v''' to represent latent vectors in R<sup>c</sup> and R<sup>v</sup><br />
<br />
<br />
''' Generative Multi-view Model: '''<br />
<br />
Consider two prior distributions over the content and view factors denoted as <math>p_c</math> and <math>p_v</math>, corresponding to the prior distribution over content and latent factors. Moreover, we consider a generator G that implements a distribution over samples x, denoted as <math>p_G</math> by computing G(c, v) with <math>c ∼ p_c</math> and <math>v ∼ p_v</math>. The objective is to learn this generator so that its first input c corresponds to the content of the generated sample while its second input v, captures the underlying view of the sample. Doing so would allow one to control the output sample of the generator by tuning its content or its view (i.e. c and v).<br />
<br />
The key idea that authors propose is to focus on the distribution of pairs of inputs rather than on the distribution over individual samples. When no view supervision is available the only valuable pairs of samples that one may build from the dataset consist of two samples of a given object under two different views. When we choose any two samples randomly from the dataset from the same object, it is most likely that we get two different views. The paper explains that there are three goals here, (i) As in regular GAN, each sample generated by G needs to look realistic. (ii) As real pairs are composed of two views of the same object, the generator should generate pairs of the same object. Since the two sampled view factors v1 and v2 are different, the only way this can be achieved is by encoding the content vector c which is invariant. (iii) It is expected that the discriminator should easily discriminate between a pair of samples corresponding to the same object under different views from a pair of samples corresponding to a same object under the same view. Because the pair shares the same content factor c, this should force the generator to use the view factors v1 and v2 to produce diversity in the generated pair.<br />
<br />
Now, the objective function of GMV Model is:<br />
<br />
<div style="text-align: center;font-size:100%"><math>\underset{G}{min} \ \underset{D}{max}</math> <math>E_{x_1,x_2}[log D(x_1,x_2)] + E_{v_1,v_2}[log(1 − D(G(c,v_1),G(c,v_2)))]</math></div><br />
<br />
Once the model is learned, generator G that generates single samples by first sampling c and v following <math>p_c</math> and <math>p_v</math>, then by computing G(c, v). By freezing c or v, one may then generate samples corresponding to multiple views of any particular content, or corresponding to many contents under a particular view. One can also make interpolations between two given views over a particular content, or between two contents using a particular view<br />
<br />
<div style="text-align: center;font-size:100%">[[File:GMV.png]]</div><br />
<br />
===Conditional Generative Model (C-GMV)===<br />
<br />
C-GMV is proposed by the authors to be able to change the view of a given object that would be provided as an input to the model. This model extends the generative model's the ability to extract the content factor from any given input and to use this extracted content in order to generate new views of the corresponding object. To achieve such a goal, we must add to our generative model an encoder function denoted <math>E : X → R^C</math> that will map any input in X to the content space <math>R^C</math><br />
<br />
Input sample x is encoded in the content space using an encoder function, noted E (implemented as a neural network).<br />
This encoder serves to generate a content vector c = E(x) that will be combined with a randomly sampled view <math>v ∼ p_v</math> to generate an artificial example. The artificial sample is then combined with the original input x to form a negative pair. The issue with this approach is that CGAN is known to easily miss modes of the underlying distribution. The generator enters in a state where it ignores the noisy component v. To overcome this phenomenon, we use the same idea as in GMV. We build negative pairs <math>(G(c, v_1), G(c, v_2))</math> by randomly sampling two views <math>v_1</math> and <math>v_2</math> that are combined to get a unique content c. c is computed from a sample x using the encoder E, i.e. c= E(x). By doing so, the ability of our approach to generating pairs with view diversity is preserved. Since this diversity can only be captured by taking into account the two different view vectors provided to the model (<math>v_1</math> and <math>v_2</math>), this will encourage G(c, v) to generate samples containing both the content information c, and the view v. Positive pairs are sampled from the training set and correspond to two views of a given object.<br />
<br />
The Objective function for C-GMV will be:<br />
<br />
<div style="text-align: center;font-size:100%"><math>\underset{G}{min} \ \underset{D}{max}</math> <math>E_{x_1,x_2 ~ p_x|l(x_1)=l(x_2)}[log D(x_1,x_2)] + E_{v_1,v_2 ~ p_v,x~p_x}[log(1 − D(G(E(x),v_1),G(E(x),v_2)))]+E_{v∼p_v,x∼p_x}[log(1 − D(G(E(x), v), x))] </math></div><br />
<br />
<div style="text-align: center;font-size:100%">[[File:CGMV.png]]</div><br />
<br />
<br />
At inference time, as with the GMV model, we are interested in getting the encoder E and the<br />
generator G. These models may be used for generating new views of any object which is observed<br />
as an input sample x by computing its content vector E(x), then sampling <math>v ∼ p_v</math> and finally by<br />
computing the output G(E(x), v)<br />
<br />
==Experiments and Results==<br />
<br />
The authors have given an exhaustive set of results and experiments.<br />
<br />
Datasets: The two models were evaluated by performing experiments over four image datasets of various domains. Note that when supervision is available on the views (like CelebA for example where images are labeled with attributes) it is not used for learning models. The only supervision that is used is if two samples correspond to the same object or not.<br />
<br />
<div style="text-align: center;font-size:100%">[[File:table_data.png]]</div><br />
<br />
<br />
Model Architecture: Same architectures for every dataset. The images were rescaled to 3×64×64 tensors. The generator G and the discriminator D follow that of the DCGAN implementation proposed in Radford et al. (2015). The encoder E is similar to D with the only differences being the batch-normalization in the first layer and the last layer which doesn't have a non-linearity. The Adam optimizer was used, with a batch size of 128. The learning rates for G and D were set to 1*10<sup>-3</sup> and 2*10<sup>-4</sup> respectively for the GMV experiments. In the C-GMV experiments, learning rates of 5*10<sup>-5</sup> were used. Alternating gradient descent was used to optimize the different objectives of the network components (generator, encoder and discriminator).<br />
<br />
Baselines: Most existing methods are learned on datasets with view labeling. To fairly compare with alternative models, authors have built baselines working in the same conditions as the models in this paper. In addition, models are compared with the model from Mathieu et al. (2016). Results gained with two implementations are reported, the first one based on the implementation provided by the authors2 (denoted Mathieu et al. (2016)), and the second one (denoted Mathieu et al. (2016) (DCGAN) ) that implements the same model using architectures inspired from DCGAN Radford et al. (2015), which is more stable and that was tuned to allow a fair comparison with our approach. For pure multi-view generative setting, generative model(GMV) is compared with standard GANs that are learned to approximate the joint generation of multiple samples: DCGANx2 is learned to output pairs of views over the same object, DCGANx4 is trained on quadruplets, and DCGANx8 on eight different views. <br />
<br />
===Generating Multiple Contents and Views===<br />
<br />
Figure 1 shows examples of generated images by our model and Figure 4 shows images sampled by the DCGAN based models (DCGANx2, DCGANx4, and DCGANx8) on 3DChairs and CelebA datasets.<br />
<br />
<div style="text-align: center;font-size:100%">[[File:fig1_gmv.png]]</div><br />
<br />
<div style="text-align: center;font-size:100%">[[File:fig4_gmv.png]]</div><br />
<br />
<br />
Figure 5 shows additional results, using the same presentation, for the GMV model only on two other datasets. In the left hand block of Figure 5, each row shows different views generated given the same content. <br />
<br />
<div style="text-align: center;font-size:100%">[[File:fig5_gmv.png]]</div><br />
<br />
Figure 6 shows generated samples obtained by interpolation between two different view factors (left) or two content factors (right). Again, in the left and right hand block of Figure 6, each row shows different views generated given the same content. It allows us to have a better idea of the underlying view/content structure captured by GMV. We can see that our approach is able to smoothly move from one content/view to another content/view while keeping the other factor constant. This also illustrates that content and view factors are well independently handled by the generator i.e. changing the view<br />
does not modify the content and vice versa.<br />
<br />
<br />
<div style="text-align: center;font-size:100%">[[File:fig6_gmv.png]]</div><br />
<br />
===Generating Multiple Views of a Given Object===<br />
<br />
The second set of experiments evaluates the ability of C-GMV to capture a particular content from an input sample and to use this content to generate multiple views of the same object. Figure 7 and 8 illustrate the diversity of views in samples generated by our model and compare our results with those obtained with the CGAN model and to models from Mathieu et al. (2016). For each row, the input sample is shown in the left column. New views are generated from that input and shown to the right, with those generated from C_GMV in the centre, and those generated from CGAN on the far right.<br />
<br />
<div style="text-align: center;font-size:100%">[[File:fig7_gmv.png]]</div><br />
<br />
<br />
<div style="text-align: center;font-size:100%">[[File:fig8_gmv.png]]</div><br />
<br />
=== Evaluation of the Quality of Generated Samples ===<br />
<br />
There are usually several metrics to evaluate generative models. Some of them are: <br />
<ol><br />
<li>Inception Score</li>: In a general sense, the Inception Score is a metric used to quantify the “realness” of a generated image. It is calculated across a set of generated images, and considers two criteria. First, all images of the sample class should be similar (low in-class variance). And second, the distribution of classes should not be dominated by any particular class. The better these criteria are met; the higher the Inception Score. <br />
<li>Latent Space Interpolation</li><br />
<li>log-likelihood (LL) score</li><br />
<li> minimum description length (MDL) score</li><br />
<li>minimum message length (MML) score</li><br />
<li>Akaike Information Criterion (AIC) score</li><br />
<li>Bayesian Information Criterion (BIC) score</li><br />
</ol><br />
<br />
<br />
<br />
<br />
<br />
The authors did sets of experiments aimed at evaluating the quality of the generated samples. They have been made on the CelebA dataset and evaluate (i) the ability of the models to preserve the identity of a person in multiple generated views, (ii) to generate realistic samples, (iii) to preserve the diversity in the generated views and (iv) to capture the view distributions of the original dataset.<br />
<br />
<div style="text-align: center;font-size:100%">[[File:tab3.png]]</div><br />
<br />
<br />
<div style="text-align: center;font-size:100%">[[File:tab4.png]]</div><br />
<br />
<br />
<div style="text-align: center;font-size:100%">[[File:table.png]]</div><br />
<br />
==Conclusion==<br />
<br />
The paper proposed a generative model, which can be learnt from multi-view data without any supervision. Moreover, it introduced a conditional version that allows generating new views of an input image. Using experiments, they proved that the model can capture content and view factors. Here, the paper showed that the application of architecture search to dense image prediction was achieved through a) The construction of a recursive search space leveraging innovation in the dense prediction literature b) construction of a fast proxy predictive of a large task. The learned architecture was shown to surpass human invented architectures across three dense image prediction tasks i.e scene parsing, person part segmentation and semantic segmentation. In the future, they are planning to use the method of this paper for data augmentation which can enrich training dataset. .<br />
<br />
==Future Work==<br />
The authors of the papers mentioned that they plan to explore using their model for data augmentation, as it can produce other data views for training, in both semi-supervised and one-shot/few-shot learning settings. <br />
<br />
==Critique==<br />
<br />
The main idea is to train the model with pairs of images with different views. It is not that clear as to what defines a view in particular. The algorithms are largely based on earlier concepts of GAN and CGAN The authors give reference to the previous papers tackling the same problem and clearly define that the novelty in this approach is not making use of view labels. The authors give a very thorough list of experiments which clearly establish the superiority of the proposed models to baselines.<br />
<br />
However, this paper only tested the model on rather constrained examples. As was observed in the results the proposed approach seems to have a high sample complexity relying on training samples covering the full range of variations for both specified and unspecified variations. Also, the proposed model does not attempt to disentangle variations within the specified and unspecified components.<br />
<br />
The method that the paper presented is novel and the paper is easy to follow. However, the authors only show a comparison between the proposed method and several baselines: DCGAN and CGAN and do not compare with the methods from Mathieu et al. 2016. In addition, the experiment result is empirical, we do not know the performance of this method in practice in the real word.<br />
<br />
==References==<br />
<br />
[1] Mickael Chen, Ludovic Denoyer, Thierry Artieres. MULTI-VIEW DATA GENERATION WITHOUT VIEW SUPERVISION. Published as a conference paper at ICLR 2018<br />
<br />
[2] Michael F Mathieu, Junbo Jake Zhao, Junbo Zhao, Aditya Ramesh, Pablo Sprechmann, and Yann LeCun. Disentangling factors of variation in deep representation using adversarial training. In Advances in Neural Information Processing Systems, pp. 5040–5048, 2016.<br />
<br />
[3] Mathieu Aubry, Daniel Maturana, Alexei Efros, Bryan Russell, and Josef Sivic. Seeing 3d chairs: exemplar part-based 2d-3d alignment using a large dataset of cad models. In CVPR, 2014.<br />
<br />
[4] Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Advances in neural information processing systems, pp. 2672–2680, 2014.<br />
<br />
[5] Emily Denton and Vighnesh Birodkar. Unsupervised learning of disentangled representations from video. arXiv preprint arXiv:1705.10915, 2017.</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=a_neural_representation_of_sketch_drawings&diff=42346a neural representation of sketch drawings2018-12-08T05:13:55Z<p>Gchalato: /* Related Work */</p>
<hr />
<div><br />
== Introduction ==<br />
In this paper, the authors present a recurrent neural network, sketch-rnn, that can be used to construct stroke-based drawings. Besides new robust training methods, they also outline a framework for conditional and unconditional sketch generation.<br />
<br />
Neural networks have been heavily used as image generation tools. For example, Generative Adversarial Networks, Variational Inference, and Autoregressive models have been used. Most of those models are designed to generate pixels to construct images. However, people learn to draw using sequences of strokes, beginning when they are young. The authors propose a new generative model that creates vector images so that it might generalize abstract concepts in a manner more similar to how humans do. <br />
<br />
The model is trained with hand-drawn sketches as input sequences. The model is able to produce sketches in vector format. In the conditional generation model, they also explore the latent space representation for vector images and discuss a few future applications of this model. The model and dataset are now available as an open source project ([https://magenta.tensorflow.org/sketch_rnn link]).<br />
<br />
=== Terminology ===<br />
Pixel images, also referred to as raster or bitmap images are files that encode image data as a set of pixels. These are the most common image type, with extensions such as .png, .jpg, .bmp. <br />
<br />
Vector images are files that encode image data as paths between points. SVG and EPS file types are used to store vector images. <br />
<br />
For a visual comparison of raster and vector images, see this [https://www.youtube.com/watch?v=-Fs2t6P5AjY video]. As mentioned, vector images are generally simpler and more abstract, whereas raster images generally are used to store detailed images. <br />
<br />
For this paper, the important distinction between the two is that the encoding of images in the model will be inherently more abstract because of the vector representation. The intuition is that generating abstract representations is more effective using a vector representation. <br />
<br />
== Related Work ==<br />
There are some works in the history that used a similar approach to generate images such as Portrait Drawing by Paul the Robot [26, 28] and some reinforcement learning approaches[28], Reinforcement Learning to discover a set of paint brush strokes that can best represent a given input photograph. They work more like a mimic of digitized photographs. There are also some Neural networks based approaches, but those are mostly dealing with pixel images. Little work is done on vector images generation. There are models that use Hidden Markov Models [25] or Mixture Density Networks [2] to generate human sketches, continuous data points (modelling Chinese characters as a sequence of pen stroke actions) or vectorized Kanji characters [9,29].<br />
<br />
Neural Network-based approaches are able to generate latent space representation of vector images, which follows a Gaussian distribution. The generated output of these networks is trained to match the Gaussian distribution by minimizing a given loss function. Using this idea, previous works attempted to generate a sequence-to-Sequence model with Variational Autoencoder to model sentences into latent space and using probabilistic program induction to model Omniglot dataset. Variational Autoencoders differ from regular encoders in that there is an intermediary “sampling step” between the encoder and decoder. Simply connecting the two would NOT guarantee that encoded parameters can be viewed as parameters of a normal distribution representing a latent space. In VAEs, the output of the encoder is physically put into an intermediary step that uses it as normal parameters, and provides a sample. In this way, the encoding is penalized as if it were the parameters of some Normal Distribution.<br />
<br />
One of the limiting factors that the authors mention in the field of generative vector drawings is the lack of availability of publicly available datasets. Previous datasets such as the Sketch data with 20k vector sketches was explored for feature extraction techniques. The Sketchy dataset consisting of 70k vector sketches along with pixel images was used for large-scale exploration of human sketches. The ShadowDraw system that used 30k raster images along with extracted vectorized features is an interactive system<br />
that predicts what a finished drawing looks like based on a set of incomplete brush strokes from the<br />
user while the sketch is being drawn. In all the cases, the datasets are comparatively small. The dataset proposed in this work uses a much larger dataset and has been made publicly available, and is one of the major contributions of this paper.<br />
<br />
== Major Contributions ==<br />
This paper makes the following major contributions: Authors outline a framework for both unconditional and<br />
conditional generation of vector images composed of a sequence of lines. The recurrent neural<br />
network-based generative model is capable of producing sketches of common objects in a vector<br />
format. The paper develops a training procedure unique to vector images to make the training more robust. The paper also made available<br />
a large dataset of hand drawn vector images to encourage further development of generative modelling<br />
for vector images, and also release an implementation of our model as an open source project<br />
<br />
== Methodology ==<br />
=== Dataset ===<br />
QuickDraw is a dataset with 50 million vector drawings collected by an online game [https://quickdraw.withgoogle.com/# Quick Draw!], where the players are required to draw objects belonging to a particular object class in less than 20 seconds. It contains hundreds of classes, each class has 70k training samples, 2.5k validation samples and 2.5k test samples.<br />
<br />
The data format of each sample is a representation of a pen stroke action event. The Origin is the initial coordinate of the drawing. The sketches are points in a list. Each point consists of 5 elements <math> (\Delta x, \Delta y, p_{1}, p_{2}, p_{3})</math> where x and y are the offset distance in x and y directions from the previous point. The parameters <math>p_{1}, p_{2}, p_{3}</math> represent three possible states in binary one-hot representation where <math>p_{1}</math> indicates the pen is touching the paper, <math>p_{2}</math> indicates the pen will be lifted from here, and <math>p_{3}</math> represents the drawing has ended.<br />
<br />
=== Sketch-RNN ===<br />
[[File:sketchfig2.png|700px|center]]<br />
<br />
The model is a Sequence-to-Sequence Variational Autoencoder(VAE). <br />
<br />
==== Encoder ====<br />
The encoder is a bidirectional RNN. The input is a sketch sequence denoted by <math>S =\{S_0, S_1, ... S_{N_{s}}\}</math> and a reversed sketch sequence denoted by <math>S_{reverse} = \{S_{N_{s}},S_{N_{s}-1}, ... S_0\}</math>. The final hidden layer representations of the two encoded sequences <math>(h_{ \rightarrow}, h_{ \leftarrow})</math> are concatenated to form a latent vector, <math>h</math>, of size <math>N_{z}</math>,<br />
<br />
\begin{split}<br />
&h_{ \rightarrow} = encode_{ \rightarrow }(S), \\<br />
&h_{ \leftarrow} = encode_{ \leftarrow }(S_{reverse}), \\<br />
&h = [h_{\rightarrow}; h_{\leftarrow}].<br />
\end{split}<br />
<br />
Then the authors project <math>h</math> into two vectors <math>\mu</math> and <math>\hat{\sigma}</math> of size <math>N_{z}</math>. The projection is performed using a fully connected layer. These two vectors are the parameters of the latent space Gaussian distribution that will estimate the distribution of the input data. Because standard deviations cannot be negative, an exponential function is used to convert it to all positive values. Next, a random variable with mean <math>\mu</math> and standard deviation <math>\sigma</math> is constructed by scaling a normalized IID Gaussian, <math>\mathcal{N}(0,I)</math>, <br />
<br />
\begin{split}<br />
& \mu = W_\mu h + b_\mu, \\<br />
& \hat \sigma = W_\sigma h + b_\sigma, \\<br />
& \sigma = exp( \frac{\hat \sigma}{2}), \\<br />
& z = \mu + \sigma \odot \mathcal{N}(0,I). <br />
\end{split}<br />
<br />
<br />
Note that <math>z</math> is not deterministic but a random vector that can be conditioned on an input sketch sequence.<br />
<br />
==== Decoder ====<br />
The decoder is an autoregressive RNN. The initial hidden and cell states are generated using <math>[h_0;c_0] = \tanh(W_z z + b_z)</math>. Here, <math>c_0</math> is utilized if applicable (eg. if an LSTM decoder is used). <math>S_0</math> is defined as <math>(0,0,1,0,0)</math> (the pen is touching the paper at location 0, 0). <br />
<br />
For each step <math>i</math> in the decoder, the input <math>x_i</math> is the concatenation of the previous point <math>S_{i-1}</math> and the latent vector <math>z</math>. The outputs of the RNN decoder <math>y_i</math> are parameters for a probability distribution that will generate the next point <math>S_i</math>. <br />
<br />
The authors model <math>(\Delta x,\Delta y)</math> as a Gaussian mixture model (GMM) with <math>M</math> normal distributions and model the ground truth data <math>(p_1, p_2, p_3)</math> as a categorical distribution <math>(q_1, q_2, q_3)</math> where <math>q_1, q_2\ \text{and}\ q_3</math> sum up to 1,<br />
<br />
\begin{align*}<br />
p(\Delta x, \Delta y) = \sum_{j=1}^{M} \Pi_j \mathcal{N}(\Delta x,\Delta y | \mu_{x,j}, \mu_{y,j}, \sigma_{x,j},\sigma_{y,j}, \rho _{xy,j}), where \sum_{j=1}^{M}\Pi_j = 1<br />
\end{align*}<br />
<br />
Where <math>\mathcal{N}(\Delta x,\Delta y | \mu_{x,j}, \mu_{y,j}, \sigma_{x,j},\sigma_{y,j}, \rho _{xy,j})</math> is a bi-variate Normal Distribution, with parameters means <math>\mu_x, \mu_y</math>, standard deviations <math>\sigma_x, \sigma_y</math> and correlation parameter <math>\rho_{xy}</math>. There are <math>M</math> such distributions. <math>\Pi</math> is a categorical distribution vector of length <math>M</math>. Collectively these form the mixture weights of the Gaussian Mixture model.<br />
<br />
The output vector <math>y_i</math> is generated using a fully-connected forward propagation in the hidden state of the RNN.<br />
<br />
\begin{split}<br />
&x_i = [S_{i-1}; z], \\<br />
&[h_i; c_i] = forward(x_i,[h_{i-1}; c_{i-1}]), \\<br />
&y_i = W_y h_i + b_y, \\<br />
&y_i \in \mathbb{R}^{6M+3}. \\<br />
\end{split}<br />
<br />
The output consists the probability distribution of the next data point.<br />
<br />
\begin{align*}<br />
[(\hat\Pi_1\ \mu_x\ \mu_y\ \hat\sigma_x\ \hat\sigma_y\ \hat\rho_{xy})_1\ (\hat\Pi_1\ \mu_x\ \mu_y\ \hat\sigma_x\ \hat\sigma_y\ \hat\rho_{xy})_2\ ...\ (\hat\Pi_1\ \mu_x\ \mu_y\ \hat\sigma_x\ \hat\sigma_y\ \hat\rho_{xy})_M\ (\hat{q_1}\ \hat{q_2}\ \hat{q_3})] = y_i<br />
\end{align*}<br />
<br />
<math>\exp</math> and <math>\tanh</math> operations are applied to ensure that the standard deviations are non-negative and the correlation value is between -1 and 1.<br />
<br />
\begin{align*}<br />
\sigma_x = \exp (\hat \sigma_x),\ <br />
\sigma_y = \exp (\hat \sigma_y),\ <br />
\rho_{xy} = \tanh(\hat \rho_{xy}). <br />
\end{align*}<br />
<br />
Categorical distribution probabilities for <math>(p_1, p_2, p_3)</math> using <math>(q_1, q_2, q_3)</math> can be obtained as :<br />
<br />
\begin{align*}<br />
q_k = \frac{\exp{(\hat q_k)}}{ \sum\nolimits_{j = 1}^{3} \exp {(\hat q_j)}},<br />
k \in \left\{1,2,3\right\}, <br />
\Pi _k = \frac{\exp{(\hat \Pi_k)}}{ \sum\nolimits_{j = 1}^{M} \exp {(\hat \Pi_j)}},<br />
k \in \left\{1,...,M\right\}.<br />
\end{align*}<br />
<br />
It is hard for the model to decide when to stop drawing because the probabilities of the three events <math>(p_1, p_2, p_3)</math> are very unbalanced. Researchers in the past have used different weights for each pen event probability, but the authors found this approach lacking elegance and inadequate. They define a hyperparameter representing the max length of the longest sketch in the training set denoted by <math>N_{max}</math>, and set the <math>S_i</math> to be <math>(0, 0, 0, 0, 1)</math> for <math>i > N_s</math>.<br />
<br />
The outcome sample <math>S_i^{'}</math> can be generated in each time step during sample process and fed as input for the next time step. The process will stop when <math>p_3 = 1</math> or <math>i = N_{max}</math>. The output is not deterministic but conditioned random sequences. The level of randomness can be controlled using a temperature parameter <math>\tau</math>.<br />
<br />
\begin{align*}<br />
\hat q_k \rightarrow \frac{\hat q_k}{\tau}, <br />
\hat \Pi_k \rightarrow \frac{\hat \Pi_k}{\tau}, <br />
\sigma_x^2 \rightarrow \sigma_x^2\tau, <br />
\sigma_y^2 \rightarrow \sigma_y^2\tau. <br />
\end{align*}<br />
<br />
The <math>\tau</math> ranges from 0 to 1. When <math>\tau = 0</math> the output will be deterministic as the sample will consist of the points on the peak of the probability density function.<br />
<br />
=== Unconditional Generation ===<br />
There is a special case that only the decoder RNN module is trained. The decoder RNN could work as a standalone autoregressive model without latent variables. In this case, initial states are 0, the input <math>x_i</math> is only <math>S_{i-1}</math> or <math>S_{i-1}^{'}</math>. In the Figure 3, generating sketches unconditionally from the temperature parameter <math>\tau = 0.2</math> at the top in blue, to <math>\tau = 0.9</math> at the bottom in red.<br />
<br />
[[File:sketchfig3.png|700px|center]]<br />
<br />
=== Training ===<br />
The training process is the same as a Variational Autoencoder. The loss function is the sum of Reconstruction Loss <math>L_R</math> and the Kullback-Leibler Divergence Loss <math>L_{KL}</math>. The reconstruction loss <math>L_R</math> can be obtained with generated parameters of pdf and training data <math>S</math>. It is the sum of the <math>L_s</math> and <math>L_p</math>, which are the log loss of the offset <math>(\Delta x, \Delta y)</math> and the pen state <math>(p_1, p_2, p_3)</math>.<br />
<br />
\begin{align*}<br />
L_s = - \frac{1 }{N_{max}} \sum_{i = 1}^{N_s} \log(\sum_{i = 1}^{M} \Pi_{j,i} \mathcal{N}(\Delta x,\Delta y | \mu_{x,j,i}, \mu_{y,j,i}, \sigma_{x,j,i},\sigma_{y,j,i}, \rho _{xy,j,i})), <br />
\end{align*}<br />
\begin{align*}<br />
L_p = - \frac{1 }{N_{max}} \sum_{i = 1}^{N_{max}} \sum_{k = 1}^{3} p_{k,i} \log (q_{k,i}), <br />
L_R = L_s + L_p.<br />
\end{align*}<br />
<br />
<br />
Both terms are normalized by <math>N_{max}</math>.<br />
<br />
<math>L_{KL}</math> measures the difference between the distribution of the latent vector <math>z</math> and an i.i.d. Gaussian vector with zero mean and unit variance.<br />
<br />
\begin{align*}<br />
L_{KL} = - \frac{1}{2 N_z} (1+\hat \sigma - \mu^2 - \exp(\hat \sigma))<br />
\end{align*}<br />
<br />
The overall loss is weighted as:<br />
<br />
\begin{align*}<br />
Loss = L_R + w_{KL} L_{KL}<br />
\end{align*}<br />
<br />
When <math>w_{KL} = 0</math>, the model becomes a standalone unconditional generator. Specially, there will be no <math>L_{KL} </math> term as we only optimize for <math>L_{R} </math>. By removing the <math>L_{KL} </math> term the model approaches a pure autoencoder, meaning it sacrifices the ability to enforce a prior over the latent space and gains better reconstruction loss metrics.<br />
<br />
While the aforementioned loss function could be used, it was found that annealing the KL term (as shown below) in the loss function produces better results.<br />
<br />
<center><math><br />
\eta_{step} = 1 - (1 - \eta_{min})R^{step}<br />
</math></center><br />
<br />
<center><math><br />
Loss_{train} = L_R + w_{KL} \eta_{step} max(L_{KL}, KL_{min})<br />
</math></center><br />
<br />
As shown in Figure 4, the <math>L_{R} </math> metric for the standalone decoder model is actually an upper bound for different models using a latent vector. The reason is the unconditional model does not access to the entire sketch it needs to generate.<br />
<br />
[[File:s.png|600px|thumb|center|Figure 4. Tradeoff between <math>L_{R} </math> and <math>L_{KL} </math>, for two models trained on single class datasets (left).<br />
Validation Loss Graph for models trained on the Yoga dataset using various <math>w_{KL} </math>. (right)]]<br />
<br />
== Experiments ==<br />
The authors experiment with the sketch-rnn model using different settings and recorded both losses. They used a Long Short-Term Memory(LSTM) model as an encoder and a HyperLSTM as a decoder. HyperLSTM is a type of RNN cell that excels at sequence generation tasks. The ability for HyperLSTM to spontaneously augment its own weights enables it to adapt to many different regimes<br />
in a large diverse dataset. They also conduct multi-class datasets. The result is as follows.<br />
<br />
[[File:sketchtable1.png|700px|center]]<br />
<br />
We could see the trade-off between <math>L_R</math> and <math>L_{KL}</math> in this table clearly. Furthermore, <math>L_R</math> decreases as <math>w_{KL} </math> is halfed. <br />
<br />
=== Conditional Reconstruction ===<br />
The authors assess the reconstructed sketch with a given sketch with different <math>\tau</math> values. We could see that with high <math>\tau</math> value on the right, the reconstructed sketches are more random.<br />
<br />
[[File:sketchfig5.png|700px|center]]<br />
<br />
They also experiment on inputting a sketch from a different class. The output will still keep some features from the class that the model is trained on.<br />
<br />
=== Latent Space Interpolation ===<br />
The authors visualize the reconstruction sketches while interpolating between latent vectors using different <math>w_{KL}</math> values. With high <math>w_{KL}</math> values, the generated images are more coherently interpolated.<br />
<br />
[[File:sketchfig6.png|700px|center]]<br />
<br />
=== Sketch Drawing Analogies ===<br />
Since the latent vector <math>z</math> encode conceptual features of a sketch, those features can also be used to augment other sketches that do not have these features. This is possible when models are trained with low <math>L_{KL}</math> values. The authors are able to perform vector arithmetic on latent vectors from different sketches and explore how the model generates sketches base on these latent spaces.<br />
<br />
=== Predicting Different Endings of Incomplete Sketches === <br />
This model is able to predict an incomplete sketch by encoding the sketch into hidden state <math>h</math> using the decoder and then using <math>h</math> as an initial hidden state to generate the remaining sketch. The authors train on individual classes by using decoder-only models and set <math>τ = 0.8</math> to complete samples. Figure 7 shows the results.<br />
<br />
[[File:sketchfig7.png|700px|center]]<br />
<br />
== Limitations ==<br />
<br />
Although sketch-rnn can model a large variety of sketch drawings, there are several limitations in the current approach. For most single-class datasets, sketch-rnn is capable of modelling around 300 data points. The model becomes increasingly difficult to train beyond this length. For the author's dataset, the Ramer-Douglas-Peucker algorithm is used to simplify the strokes of sketch data to less than 200 data points.<br />
<br />
For more complicated classes of images, such as mermaids or lobsters, the reconstruction loss metrics are not as good compared to simpler classes such as ants, faces or firetrucks. The models trained on these more challenging image classes tend to draw smoother, more circular line segments that do not resemble individual sketches, but rather resemble an averaging of many sketches in the training set. This smoothness may be analogous to the blurriness effect produced by a Variational Autoencoder that is trained on pixel images. Depending on the use case of the model, smooth circular lines can be viewed as aesthetically pleasing and a desirable property.<br />
<br />
While both conditional and unconditional models are capable of training on datasets of several classes, sketch-rnn is ineffective at modelling a large number of classes simultaneously. The samples generated will be incoherent, with different classes are shown in the same sketch.<br />
<br />
== Applications and Future Work ==<br />
The authors believe this model can assist artists by suggesting how to finish a sketch, helping them to find interesting intersections between different drawings or objects, or generating a lot of similar but different designs. In the simplest use, pattern designers can apply sketch-rnn to generate a large number of similar, but unique designs for textile or wallpaper prints. The creative designers can also come up with abstract designs which enables them to resonate more with their target audience<br />
<br />
This model may also find its place on teaching students how to draw. Even with the simple sketches in QuickDraw, the authors of this work have become much more proficient at drawing animals, insects, and various sea creatures after conducting these experiments. <br />
When the model is trained with a high <math>w_{KL}</math> and sampled with a low <math>\tau</math>, it may help to turn a poor sketch into a more aesthetical one. Latent vector augmentation could also help to create a better drawing by inputting user-rating data during training processes.<br />
<br />
The authors conclude by providing the following future directions to this work:<br />
# Investigate using user-rating data to augmenting the latent vector in the direction that maximizes the aesthetics of the drawing.<br />
# Look into combining variations of sequence-generation models with unsupervised, cross-domain pixel image generation models.<br />
<br />
It's exciting that they manage to combine this model with other unsupervised, cross-domain pixel image generation models to create photorealistic images from sketches.<br />
<br />
The authors have also mentioned the opposite direction of converting a photograph of an object into an unrealistic, but similar looking<br />
sketch of the object composed of a minimal number of lines to be a more interesting problem.<br />
<br />
Moreover, it would be interesting to see how varying loss will be represented as a drawing. Some exotic form of loss function may change the way that the network behaves, which can lead to various applications.<br />
<br />
== Conclusion ==<br />
The paper presents a methodology to model sketch drawings using recurrent neural networks. The sketch-rnn model that can encode and decode sketches, generate and complete unfinished sketches is introduced in this paper. In addition, Authors demonstrated how to both interpolate between latent spaces from a different class, and use it to augment sketches or generate similar looking sketches. Furthermore, the importance of enforcing a prior distribution on latent vector while interpolating coherent sketch generations is shown. Finally, a large sketch drawings dataset for future research work is created.<br />
<br />
== Critique ==<br />
This paper presents both a novel large dataset of sketches and a new RNN architecture to generate new sketches. It is very exciting to read but there are still some aspect to improve.<br />
<br />
* The performance of the decoder model can hardly be evaluated. The authors present the performance of the decoder by showing the generated sketches, it is clear and straightforward, however, not very efficient. It would be great if the authors could present a way, or a metric to evaluate how well the sketches are generated rather than printing them out and evaluate with human judgment. The authors didn't present an evaluation of the algorithms either. They provided <math>L_R</math> and <math>L_{KL}</math> for reference, however, a lower loss doesn't represent a better performance. Training loss alone likely does not capture the quality of a sketch.<br />
<br />
* The authors have not mentioned details on training details such as learning rate, training time, parameter size, and so on. <br />
<br />
* Algorithm lacks comparison to the prior state of the art on standard metrics, which made the novelty unclear. Using strokes as inputs is a novel and innovative move, however, the paper does not provide a baseline or any comparison with other methods or algorithms. Some other researches were mentioned in the paper, using similar and smaller datasets. It would be great if the authors could use some basic or existing methods a baseline and compare with the new algorithm.<br />
<br />
* Besides the comparison with other algorithms, it would also be great if the authors could remove or replace some component of the algorithm in the model to show if one part is necessary, or what made them decide to include a specific component in the algorithm.<br />
<br />
* The authors did not present better complexity and deeper mathematical analysis on the algorithms in the paper. It also does not include comparison using some more standard metrics compare to previous results. Therefore, it lacks some algorithmic contribution. It would be better to include some more formal analysis on the algorithmic side. <br />
<br />
* The authors proposed a few future applications for the model, however, the current output seems somehow not very close to their descriptions. But I do believe that this is a very good beginning, with the release of the sketch dataset, it must attract more scholars to research and improve with it!<br />
<br />
* As they said their model can become increasingly difficult to train on with increased size.<br />
<br />
== References == <br />
# Jimmy L. Ba, Jamie R. Kiros, and Geoffrey E. Hinton. Layer normalization. NIPS, 2016.<br />
# Christopher M. Bishop. Mixture density networks. Technical Report, 1994. URL http://publications.aston.ac.uk/373/.<br />
# Samuel R. Bowman, Luke Vilnis, Oriol Vinyals, Andrew M. Dai, Rafal Józefowicz, and Samy Bengio. Generating Sentences from a Continuous Space. CoRR, abs/1511.06349, 2015. URL http://arxiv.org/abs/1511.06349.<br />
# H. Dong, P. Neekhara, C. Wu, and Y. Guo. Unsupervised Image-to-Image Translation with Generative Adversarial Networks. ArXiv e-prints, January 2017.<br />
# David H. Douglas and Thomas K. Peucker. Algorithms for the reduction of the number of points required to represent a digitized line or its caricature. Cartographica: The International Journal for Geographic Information and Geovisualization, 10(2):112–122, October 1973. doi: 10.3138/fm57-6770-u75u-7727. URL http://dx.doi.org/10.3138/fm57-6770-u75u-7727.<br />
# Mathias Eitz, James Hays, and Marc Alexa. How Do Humans Sketch Objects? ACM Trans. Graph.(Proc. SIGGRAPH), 31(4):44:1–44:10, 2012.<br />
# I. Goodfellow. NIPS 2016 Tutorial: Generative Adversarial Networks. ArXiv e-prints, December 2016.<br />
# Alex Graves. Generating sequences with recurrent neural networks. arXiv:1308.0850, 2013.<br />
# David Ha. Recurrent Net Dreams Up Fake Chinese Characters in Vector Format with TensorFlow, 2015.<br />
# David Ha, Andrew M. Dai, and Quoc V. Le. HyperNetworks. In ICLR, 2017.<br />
# Sepp Hochreiter and Juergen Schmidhuber. Long short-term memory. Neural Computation, 1997.<br />
# P. Isola, J.-Y. Zhu, T. Zhou, and A. A. Efros. Image-to-Image Translation with Conditional Adversarial Networks. ArXiv e-prints, November 2016.<br />
# Jonas Jongejan, Henry Rowley, Takashi Kawashima, Jongmin Kim, and Nick Fox-Gieg. The Quick, Draw! - A.I. Experiment. https://quickdraw.withgoogle.com/, 2016. URL https: //quickdraw.withgoogle.com/.<br />
# C. Kaae Sønderby, T. Raiko, L. Maaløe, S. Kaae Sønderby, and O. Winther. Ladder Variational Autoencoders. ArXiv e-prints, February 2016.<br />
# T. Kim, M. Cha, H. Kim, J. Lee, and J. Kim. Learning to Discover cross-domain Relations with Generative Adversarial Networks. ArXiv e-prints, March 2017.<br />
# D. P Kingma and M. Welling. Auto-Encoding Variational Bayes. ArXiv e-prints, December 2013.<br />
# Diederik Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In ICLR, 2015.<br />
# Diederik P. Kingma, Tim Salimans, and Max Welling. Improving variational inference with inverse autoregressive flow. CoRR, abs/1606.04934, 2016. URL http://arxiv.org/abs/1606.04934.<br />
# Brenden M. Lake, Ruslan Salakhutdinov, and Joshua B. Tenenbaum. Human level concept learning through probabilistic program induction. Science, 350(6266):1332–1338, December 2015. ISSN 1095-9203. doi: 10.1126/science.aab3050. URL http://dx.doi.org/10.1126/science.aab3050.<br />
# Yong Jae Lee, C. Lawrence Zitnick, and Michael F. Cohen. Shadowdraw: Real-time user guidance for freehand drawing. In ACM SIGGRAPH 2011 Papers, SIGGRAPH ’11, pp. 27:1–27:10, New York, NY, USA, 2011. ACM. ISBN 978-1-4503-0943-1. doi: 10.1145/1964921.1964922. URL http://doi.acm.org/10.1145/1964921.1964922.<br />
# M.-Y. Liu, T. Breuel, and J. Kautz. Unsupervised Image-to-Image Translation Networks. ArXiv e-prints, March 2017.<br />
# S. Reed, A. van den Oord, N. Kalchbrenner, S. Gómez Colmenarejo, Z. Wang, D. Belov, and N. de Freitas. Parallel Multiscale Autoregressive Density Estimation. ArXiv e-prints, March 2017.<br />
# Patsorn Sangkloy, Nathan Burnell, Cusuh Ham, and James Hays. The Sketchy Database: Learning to Retrieve Badly Drawn Bunnies. ACM Trans. Graph., 35(4):119:1–119:12, July 2016. ISSN 0730-0301. doi: 10.1145/2897824.2925954. URL http://doi.acm.org/10.1145/2897824.2925954.<br />
# Mike Schuster, Kuldip K. Paliwal, and A. General. Bidirectional recurrent neural networks. IEEE Transactions on Signal Processing, 1997.<br />
# Saul Simhon and Gregory Dudek. Sketch interpretation and refinement using statistical models. In Proceedings of the Fifteenth Eurographics Conference on Rendering Techniques, EGSR’04, pp. 23–32, Aire-la-Ville, Switzerland, Switzerland, 2004. Eurographics Association. ISBN 3-905673-12-6. doi: 10.2312/EGWR/EGSR04/023-032. URL http://dx.doi.org/10.2312/EGWR/EGSR04/023-032.<br />
# Patrick Tresset and Frederic Fol Leymarie. Portrait drawing by paul the robot. Comput. Graph.,37(5):348–363, August 2013. ISSN 0097-8493. doi: 10.1016/j.cag.2013.01.012. URL http://dx.doi.org/10.1016/j.cag.2013.01.012.<br />
# T. White. Sampling Generative Networks. [https://arxiv.org/abs/1609.04468 ArXiv e-prints], September 2016.<br />
#Ning Xie, Hirotaka Hachiya, and Masashi Sugiyama. Artist agent: A reinforcement learning approach to automatic stroke generation in oriental ink painting. In ICML. icml.cc / Omnipress, 2012. URL http://dblp.uni-trier.de/db/conf/icml/icml2012.html#XieHS12.<br />
# Xu-Yao Zhang, Fei Yin, Yan-Ming Zhang, Cheng-Lin Liu, and Yoshua Bengio. Drawing and Recognizing Chinese Characters with Recurrent Neural Network. CoRR, abs/1606.06539, 2016. URL http://arxiv.org/abs/1606.06539.</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=DETECTING_STATISTICAL_INTERACTIONS_FROM_NEURAL_NETWORK_WEIGHTS&diff=42345DETECTING STATISTICAL INTERACTIONS FROM NEURAL NETWORK WEIGHTS2018-12-08T05:05:42Z<p>Gchalato: /* Notations */</p>
<hr />
<div>=Introduction=<br />
<br />
It has been commonly believed that one major advantage of neural networks is their capability of modelling complex statistical interactions between features for automatic feature learning. Statistical interactions capture important information on where features often have joint effects with other features on predicting an outcome. The discovery of interactions is especially useful for scientific discoveries and hypothesis validation. For example, physicists may be interested in understanding what joint factors provide evidence for new elementary particles; doctors may want to know what interactions are accounted for in risk prediction models, to compare against known interactions from existing medical literature.<br />
<br />
With the growth in the computational power available Neural Networks have been able to solve many of the complex tasks in a wide variety of fields. This is mainly due to their ability to model complex and non-linear interactions. Neural networks have traditionally been treated as “black box” models, preventing their adoption in many application domains, such as those where explainability is desirable. It has been noted that complex machine learning models can learn unintended patterns from data, raising significant risks to stakeholders [14]. Therefore, in applications where machine learning models are intended for making critical decisions, such as healthcare or finance, it is paramount to understand how they make predictions [9]. Within several areas, like eg: computation social science, interpretability is of utmost importance. Since we do not understand how a neural network comes to its decision, practitioners in these areas tend to prefer simpler models like linear regression, decision trees, etc. which are much more interpretable. In this paper, we are going to present one way of implementing interpretability in a neural network.<br />
<br />
Existing approaches to interpreting neural networks can be summarized into two types. One type is direct interpretation, which focuses on 1) explaining individual feature importance, for example by computing input gradients [13] and decomposing predictions [8], 2) developing attention-based models, which illustrate where neural networks focus during inference [11], and 3) providing model-specific visualizations, such as feature map and gate activation visualizations [15]. The other type is indirect interpretation, for example post-hoc interpretations of feature importance [12] and knowledge distillation to simpler interpretable models [10].<br />
<br />
In this paper, the authors propose Neural Interaction Detection (NID), which can detect any order or form of statistical interaction captured by the feedforward neural network by examining its weight matrix. This approach is efficient because it avoids searching over an exponential solution space of interaction candidates by making an approximation of hidden unit importance at the first hidden layer via all weights above and doing a 2D traversal of the input weight matrix.<br />
<br />
Note that in this paper, we only consider one specific types of neural network, feedforward neural network. Based on the methodology discussed here, the authors suggest that we can build an interpretation method for other types of networks also.<br />
<br />
=Related Work=<br />
<br />
1. Interaction Detection approaches: <br />
* Conduct individual tests for all features' combination such as ANOVA and Additive Groves. Two-way ANOVA has been a standard method of performing pairwise interaction detection that involves conducting hypothesis tests for each interaction candidate by checking each hypothesis with F-statistics (Wonnacott & Wonnacott, 1972). Additive Groves is another method that conducts individual tests for interactions and hence must face the same computational difficulties; however, it is special because the interactions it detects are not constrained to any functional form.<br />
* Define all interaction forms of interest, then later finds the important ones.<br />
<br />
- The paper's goal is to detect interactions without compromising the functional forms. Our method accomplishes higher-order interaction detection, which has the benefit of avoiding a high false positive or false discovery rate.<br />
<br />
2. Interpretability: A lot of work has also been done in this particular area and it can be divided it the following broad categories:<br />
* Feature Importance through Decomposition: Methods like Input Gradient(Sundararajan et al., 2017) learns the importance of features through a gradient-based approach similar to backpropagation. Works like Li et al(2017), Murdoch(2017) and Murdoch(2018) study interpretability of LSTMs by looking at phrase and word level importance scores. Bach et al. 2015 and Shrikumar et al. 2016 (DeepLift) study pixel importance in CNNs.<br />
* Studying Visualizations in Models - Karpathy et al. (2015) worked with character generating LSTMs and tried to study activation and firing in certain hidden units for meaningful attributes. (Yosinski et al., 2015 studies feature map visualizations, providing a tool for visualizing live activations on each layer of a trained CNN, and another for visualizing "Regularized Optimization".) <br />
* Attention-Based Models: Bahdanau et al. (2014) - These are a different class of models which use attention modules(different architectures) to help focus the neural network to decide the parts of the input that it should look more closely or give more importance to. Looking at the results of these type of model an indirect sense of interpretability can be gauged.<br />
* Sum product networks, Hoifun Poon, Pedro Domingos (2011) It is a new deep architecture that provides clear semantics. In its core, it is a probabilistic model, with two types of nodes: Sum node and Product nodes. The sum nodes are trying to model the mixture of distributions and product node is trying to model joint distributions. It can be trained using gradient descent and other methods as well. The main advantage of the Sum-Product Network is that it has clear semantics, where people can interpret exactly how the network models make decisions. Therefore, it has better interpretability than most of the current deep architectures. <br />
<br />
The approach in this paper is to extract non-additive interactions between variables from the neural network weights.<br />
<br />
=Notations=<br />
Before we dive in to methodology, we are going to define a few notations here. Most of them will be trivial.<br />
<br />
1. Vector: Vectors are defined with bold-lowercases, '''v, w'''<br />
<br />
2. Matrix: Matrices are defined with bold-uppercases, '''V, W'''<br />
<br />
3. Interger Set: For some interger p <math>\in</math> Z, we define [p] := {1,2,3,...,p}<br />
<br />
=Interaction=<br />
First of all, in order to explain the model, we need to be able to explain the interactions and their effects to output. Therefore, we define 'interacion' between variables as below. <br />
<br />
[[File:def_interaction.PNG|900px|center]]<br />
<br />
From the definition above, for a function like, <math>x_1x_2 + sin(x_3 + x_4 + x_5)</math>, we have <math>{[x_1, x_2]}</math> and <math>{[x_3, x_4, x_5]}</math> interactions. And we say that the latter interaction to be 3-way interaction.<br />
<br />
Note that from the definition above, we can naturally deduce that d-way interaction can exist if and only if all of its (d-1) interactions exist. For example, 3-way interaction above shows that we have 2-way interactions <math>{[3,4], [4,5]}</math> and <math>{[3,5]}</math>.<br />
<br />
One thing that we need to keep in mind is that for models like neural network, most of interactions are happening within hidden layers. This means that we needa proper way of measuring interaction strength.<br />
<br />
The key observation is that for any kinds of interaction, at a some hidden unit of some hidden layer, two interacting features the ancestors. In graph-theoretical language, interaction map can be viewed as an associated directed graph and for any interaction <math>\Gamma \in [p]</math>, there exists at least one vertix that has all of features of <math>\Gamma</math> as ancestors. The statement can be rigorized as the following:<br />
<br />
<br />
[[File:prop2.PNG|900px|center]]<br />
<br />
Now, the above mathematical statement gurantees us to measure interaction strengths at ANY hidden layers. For example, if we want to study about interactions at some specific hidden layer, now we now that there exists corresponding vertices between the hidden layer and output layer. Therefore all we need to do is now to find approprite measure which can summarize the information between those two layers.<br />
<br />
Before doing so, let's think about a single-layered neural network. For any one hidden unit, we can have possibly, <math>2^{||W_i,:||}</math>, number of interactions. This means that our search space might be too huge for multi-layered networks. Therefore, we need a some descent way of approximate out search space. Moreover, the authors realized a fast interaction detection by limiting the search complexity of the task by only quantifying interactions created at the first hidden layer. The figure below illustrates an interaction within a fully connected feedforward neural network, where the box contains later layers in the network.<br />
<br />
[[File:network1.PNG|500px|center]]<br />
<br />
==Measuring influence in hidden layers==<br />
As we discussed above, in order to consider interaction between units in any layers, we need to think about their out-going paths. However, we soon encountered the fact that for some fully-connected multi-layer neural network, the search space might be too huge to compare. Therefore, we use information about out-going paths gredient upper bond. To represent the influence of out-going paths at <math>l</math>-hidden layer, we define cumulative impact of weights between output layer and <math>l+1</math>. We define aggregated weights as, <br />
<br />
[[File:def3.PNG|900px|center]]<br />
<br />
<br />
Note that <math>z^{(l)} \in R^{(p_l)}</math> where <math>p_l</math> is the number of hidden units in <math>l</math>-layer.<br />
Moreover, this is the lipschitz constant of gredients. Gredient has been an import variable of measuring influence of features, especially when we consider that input layer's derivative computes the direction normal to decision boundaries.<br />
<br />
==Quantifying influence==<br />
For some <math>i</math> hidden unit at the first hidden layer, which is the closet layer to the input layer, we define the influence strength of some interaction as, <br />
<br />
[[File:measure1.PNG|900px|center]]<br />
<br />
The function <math>\mu</math> will be defined later. Essentially, the formula shows that the strength of influence is defined as the product of the aggregated weight on the first hidden layer and some measure of influence between the first hidden layer and the input layer. <br />
<br />
For the function, <math>\mu</math>, any positive-real valued functions such as max, min and average can be candidates. The effects of those candidates will be tested later.<br />
<br />
Now based on the specifications above, the author suggested the algorithm for searching influential interactions between input layer units as follows:<br />
<br />
It was pointed out that restricting to the first hidden layer might miss some important feature interactions, however, the author state that it is not straightforward how to incorporate the idea of hidden units at intermediate layers to get better interaction detection performance.<br />
[[File:algorithm1.PNG|850px|center]]<br />
<br />
=Cut-off Model=<br />
Now using the greedy algorithm defined above, we can rank the interactions by their strength. However, in order to access true interactions, we are building the cut-off model which is a generalized additive model (GAM) as below,<br />
<br />
<center><math><br />
c_K('''x''') = \sum_{i=1}^{p}g_i(x_i) + \sum_{i=1}^{K}{g_i}^\prime(x_\chi)<br />
</math></center><br />
<br />
From the above model, each of <math>g_i</math> and <math>g_i'</math> are Feed-Forward neural networks. <math>g_i(\cdot)</math> captures the main effects, while <math>g_i'(\cdot)</math> captures the interaction. We are keep adding interactions until the performance reaches plateaus.<br />
<br />
=Experiment=<br />
For the experiment, the authors have compared three neural network model with traditional statistical interaction detecting algorithms. For the nueral network models, first model will be MLP, second model will be MLP-M, which is MLP with additional univariate network at the output. The last one is the cut-off model defined above, which is denoted by MLP-cutoff. In the experiments that the authors performed, all the networks which modelled feature interactions consisted of four hidden layers containing 140, 100, 60, and 20 units respectively. Whereas, all the individual univariate networks contained three hidden layers with each layer containing 10 units. All of these networks used ReLu activation and backpropagation for training. The MLP-M model is graphically represented below.<br />
<br />
[[File:output11.PNG|300px|center]]<br />
<br />
For the experiment, the authors study our interaction detection framework on both simulated and real-world experiments. For simulated experiments, the authors are going to test on 10 synthetic functions as shown in table I.<br />
<br />
[[File:synthetic.PNG|900px|center]]<br />
<br />
The authors use four real-world datasets, of which two are regression datasets, and the other two are binary classification datasets. The datasets are a mixture of common prediction tasks in the cal housing<br />
and bike sharing datasets, a scientific discovery task in the higgs boson dataset, and an example of very-high order interaction detection in the letter dataset.<br />
<br />
And the authors also reported the results of comparisons between the models. As you can see, neural network based models are performing better on average. Compare to the traditional methods like ANOVA, MLP and MLP-M method shows 20% increases in performance.<br />
<br />
[[File:performance_mlpm.PNG|900px|center]]<br />
<br />
<br />
[[File:performance2_mlpm.PNG|900px|center]]<br />
<br />
The above result shows that MLP-M almost perfectly capture the most influential pair-wise interactions.<br />
<br />
=Higher-order interaction detection=<br />
The authors use their greedy interaction ranking algorithm to perform higher-order interactiondetection without an exponential search of interaction candidates.<br />
[[File:higher-order_interaction_detection.png|700px|center]]<br />
<br />
=Limitations=<br />
Even though for the above synthetic experiment MLP methods showed superior performances, the method still have some limitations. For example, fir the function like, <math>x_1x_2 + x_2x_3 + x_1x_3</math>, neural network fails to distinguish between interlinked interactions to single higher order interaction. Moreoever, correlation between features deteriorates the ability of the network to distinguish interactions. However, correlation issues are presented most of interaction detection algorithms. <br />
<br />
In the case of detecting pairwise interactions, the interlinked pairwise interactions are often confused by the algorithm for complex interactions. This means that the higher-order interaction algorithm fails to separate interlinked pairwise interactions encoded in the neural network. Another issue is that it sometimes detects abrupt interactions or misses interactions as a result of correlations between features<br />
<br />
Because this method relies on the neural network fitting the data well, there are some additional concerns. Notably, if the NN is unable to make an appropriate fit (under/overfitting), the resulting interactions will be flawed. This can occur if the datasets that are too small or too noisy, which often occurs in practical settings.<br />
<br />
=Conclusion=<br />
Here we presented the method of detecting interactions using MLP. Compared to other state-of-the-art methods like Additive Groves (AG), the performances are competitive yet computational powers required is far less. Therefore, it is safe to claim that the method will be extremly useful for practitioners with (comparably) less computational powers. Moreover, the NIP algorithm successfully reduced the computation sizes. After all, the most important aspect of this algorithm is that now users of nueral networks can impose interpretability in the model usage, which will change the level of usability to another level for most of practitioners outside of those working in machine learning and deep learning areas.<br />
<br />
For future work, the authors want to detect feature interactions by using the common units in the intermediate hidden layers of feedforward networks, and also want to use such interaction detection to interpret weights in other deep neural networks. Also, it was pointed out that the neural network weights heavily depend on L-1 regularized neural network training, but a group lasso penalty may work better.<br />
<br />
=Critique=<br />
1. Authors need to do large-scale experiments, instead of just conducting experiments on some synthetic dataset with small feature dimensionality, to make their claim stronger.<br />
<br />
2. Although the method proposed in this paper is interesting, the paper would benefit from providing some more explanations to support its idea and fill the possible gaps in its experimental evaluation. In some parts there are repetitive explanations that could be replaced by other essential clarifications.<br />
<br />
3. Greedy algorithm is implemented but nothing is mentioned about the speed of this algorithm which is definitely not fast. So, this has the potential to be a weak point of the study.<br />
<br />
=Reference=<br />
<br />
[1] Jacob Bien, Jonathan Taylor, and Robert Tibshirani. A lasso for hierarchical interactions. Annals of statistics, 41(3):1111, 2013. <br />
<br />
[2] G David Garson. Interpreting neural-network connection weights. AI Expert, 6(4):46–51, 1991.<br />
<br />
[3] Yotam Hechtlinger. Interpretation of prediction models using the input gradient. arXiv preprint arXiv:1611.07634, 2016.<br />
<br />
[4] Shiyu Liang and R Srikant. Why deep neural networks for function approximation? 2016. <br />
<br />
[5] David Rolnick and Max Tegmark. The power of deeper networks for expressing natural functions. International Conference on Learning Representations, 2018. <br />
<br />
[6] Daria Sorokina, Rich Caruana, and Mirek Riedewald. Additive groves of regression trees. Machine Learning: ECML 2007, pp. 323–334, 2007.<br />
<br />
[7] Simon Wood. Generalized additive models: an introduction with R. CRC press, 2006<br />
<br />
[8] Sebastian Bach, Alexander Binder, Gre ́goire Montavon, Frederick Klauschen, Klaus-Robert Mu ̈ller, and Wojciech Samek. On pixel-wise explanations for non-linear classifier decisions by layer-wise relevance propagation. PloS one, 10(7):e0130140, 2015.<br />
<br />
[9] Rich Caruana, Yin Lou, Johannes Gehrke, Paul Koch, Marc Sturm, and Noemie Elhadad. Intel- ligible models for healthcare: Predicting pneumonia risk and hospital 30-day readmission. In Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 1721–1730. ACM, 2015.<br />
<br />
[10] Zhengping Che, Sanjay Purushotham, Robinder Khemani, and Yan Liu. Interpretable deep models for icu outcome prediction. In AMIA Annual Symposium Proceedings, volume 2016, pp. 371. American Medical Informatics Association, 2016.<br />
<br />
[11] Laurent Itti, Christof Koch, and Ernst Niebur. A model of saliency-based visual attention for rapid scene analysis. IEEE Transactions on pattern analysis and machine intelligence, 20(11):1254– 1259, 1998.<br />
<br />
[12] Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. Why should i trust you?: Explaining the predictions of any classifier. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 1135–1144. ACM, 2016.<br />
<br />
[13]Karen Simonyan, Andrea Vedaldi, and Andrew Zisserman. Deep inside convolutional networks: Vi- sualising image classification models and saliency maps. arXiv preprint arXiv:1312.6034, 2013.<br />
<br />
[14] Kush R Varshney and Homa Alemzadeh. On the safety of machine learning: Cyber-physical sys- tems, decision sciences, and data products. arXiv preprint arXiv:1610.01256, 2016.<br />
<br />
[15] Jason Yosinski, Jeff Clune, Anh Nguyen, Thomas Fuchs, and Hod Lipson. Understanding neural networks through deep visualization. arXiv preprint arXiv:1506.06579, 2015.</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Fix_your_classifier:_the_marginal_value_of_training_the_last_weight_layer&diff=42344Fix your classifier: the marginal value of training the last weight layer2018-12-08T05:00:19Z<p>Gchalato: /* Previous Work */</p>
<hr />
<div><br />
The code for the proposed model is available at https://github.com/eladhoffer/fix_your_classifier.<br />
<br />
=Introduction=<br />
<br />
Deep neural networks have become widely used for machine learning, achieving state-of-the-art results on many tasks. One of the most common tasks they are used for is classification. For example, convolutional neural networks (CNNs) are used to classify images to a semantic category. Typically, a learned affine transformation is placed at the end of such models, yielding a per-class prediction for each class. This classifier can have a vast number of parameters, which grows linearly with the number of possible classes. As the number of classes in the data set increases, more and more computational resources are consumed by this layer.<br />
<br />
=Brief Overview=<br />
<br />
In order to alleviate the aforementioned problem, the authors propose that the final layer of the classifier be fixed (up to a global scale constant). They argue that with little or no loss of accuracy for most classification tasks, the method provides significant memory and computational benefits. In addition, they show that by initializing the classifier with a Hadamard matrix the inference could be made faster as well.<br />
<br />
=Previous Work=<br />
<br />
Training neural networks (NNs) and using them for inference requires large amounts of memory and computational resources; thus, extensive research has focused on reducing the size of these networks, including:<br />
<br />
* Weight sharing and specification (Han et al., 2015). Weight sharing has to do with the filters used in the convolution layers. Normally, we would train different filters for each depth dimension. For example, in a particular convolution step, if our input was XxYx3, we would have 3 different filters applied to each 2D slice. Weight sharing forces the network to use identical filters for each of these layers. That is, the trained-for AxB convolution will have the same values, and the “stamp” will be identical; no matter which of the 3 dimensions is dealt with. <br />
<br />
* Mixed precision to reduce the size of the neural networks by half (Micikevicius et al., 2017)<br />
<br />
* Low-rank approximations to speed up CNN (Tai et al., 2015)<br />
<br />
* Quantization of weights, activations, and gradients to further reduce computation during training (Hubara et al., 2016b; Li et al., 2016 and Zhou et al., 2016). Although aggressive quantization benefits from smaller model size, the extreme compression rate comes with a loss of accuracy.<br />
<br />
Some of the past works have also put forward the fact that predefined (Park & Sandberg, 1991) and random (Huang et al., 2006) projections can be used together with a learned affine transformation to achieve competitive results on many of the classification tasks. However, the authors' proposal in the current paper suggests the reverse proposal - common NN models can learn useful representations even without modifying the final output layer, which often holds a large number of parameters that grows linearly with number of classes.<br />
<br />
=Background=<br />
<br />
A CNN is comprised of one or more convolutional layers (often with a subsampling step) and then followed by one or more fully connected layers as in a standard multilayer neural network. The architecture of a CNN is designed to take advantage of the 2D structure of an input image (or other 2D input such as a speech signal). This is achieved with local connections and tied weights followed by some form of pooling which results in translation invariant features. Another benefit of CNNs is that they are easier to train and have many fewer parameters than fully connected networks with the same number of hidden units. <br />
<br />
A CNN consists of a number of convolutional and subsampling layers optionally followed by fully connected layers. The input to a convolutional layer is a <math>m \times m \times r</math> image where <math>m</math> is the height and width of the image and <math>r</math> is the number of channels, e.g. an RGB image has <math>r=3</math>. The convolutional layer can have a specifiable number of <math>k</math> filters (or kernels) of size <math>n \times n \times q</math> where <math>n</math> is the height and width of the kernel and is smaller than the dimension of the image, and <math>q</math> can either be the same or smaller that the number of channels in the previous layer. Each kernel is convolved with the image to produce <math>k</math> feature maps of size <math>m−n+1</math>. Each map is then subsampled typically with mean or max pooling over <math>p \times p</math> contiguous regions where <math>p</math> ranges between 2 for small images (e.g. MNIST) and is usually not more than 5 for larger inputs. Either before or after the subsampling layer an additive bias and sigmoidal nonlinearity is applied to each feature map. <br />
<br />
CNNs are commonly used to solve a variety of spatial and temporal tasks. Earlier architectures of CNNs (LeCun et al., 1998; Krizhevsky et al., 2012) used a set of fully-connected layers at later stages of the network, presumably to allow classification based on global features of an image.<br />
<br />
<br />
== Shortcomings of the Final Classification Layer ==<br />
<br />
Zeiler & Fergus, 2014 showed that despite the enormous number of trainable parameters fully connected layers add to the model, they have a rather marginal impact on the performance. Furthermore, it has been shown that these layers could be easily compressed and reduced after a model is trained by simple means such as matrix decomposition and sparsification (Han et al., 2015). Modern NNs are characterized with the removal of most fully connected layers (Lin et al., 2013; Szegedy et al., 2015; He et al., 2016). This was shown to lead to better generalization and overall accuracy, together with a huge decrease in the number of trainable parameters. Additionally, numerous works showed that CNNs can be trained in a metric learning regime (Bromley et al., 1994; Schroff et al., 2015; Hoffer & Ailon, 2015), where no explicit classification layer is used and the objective regarded only distance measures between intermediate representations. Hardt & Ma (2017) suggested an all-convolutional network variant, where they kept the original initialization of the classification layer fixed and found no negative impact on performance on the CIFAR-10 dataset.<br />
<br />
=Proposed Method=<br />
<br />
The aforementioned works provide evidence that fully-connected layers are in fact redundant and play a small role in learning and generalization. In this work, the authors have suggested that the parameters used for the final classification transform are completely redundant, and can be replaced with a predetermined linear transform. This holds for even in large-scale models and classification tasks, such as recent architectures trained on the ImageNet benchmark (Deng et al., 2009).<br />
<br />
==Using a Fixed Classifier==<br />
<br />
Suppose the final representation obtained by the network (the last hidden layer) is represented as <math>x = F(z;\theta)</math> where <math>F</math> is assumed to be a deep neural network with input <math>z</math> and parameters <math>θ</math>, e.g., a convolutional network, trained by backpropagation. In common NN models, this representation is followed by an additional affine transformation, <math>y = W^T x + b</math>, where <math>W</math> and <math>b</math> are also trained by back-propagation.<br />
<br />
For input <math>x</math> of length <math>N</math>, and <math>C</math> different possible outputs, <math>W</math> is required to be a matrix of <math>N × C</math>. <br />
<br />
Training is done using cross-entropy loss, by feeding the network outputs through a softmax activation<br />
<br />
\begin{align}<br />
v_i = \frac{e^{y_i}}{\sum_{j}^{C}{e^{y_j}}}, \qquad i \in {1, . . . , C}<br />
\end{align}<br />
<br />
and reducing the expected negative log likelihood with respect to ground-truth target class <math> t </math>, by minimizing the loss function:<br />
<br />
\begin{align}<br />
L(x, t) = −\text{log}\ {v_t} = −{w_t}·{x} − b_t + \text{log} ({\sum_{j}^{C}e^{w_j . x + b_j}})<br />
\end{align}<br />
<br />
where <math>w_i</math> is the <math>i</math>-th column of <math>W</math>.<br />
<br />
==Choosing the Projection Matrix==<br />
<br />
To evaluate the conjecture regarding the importance of the final classification transformation, the trainable parameter matrix <math>W</math> is replaced with a fixed orthonormal projection <math> Q &isin; R^{N×C} </math>, such that <math> &forall; i &ne; j : q_i · q_j = 0 </math> and <math> || q_i ||_{2} = 1 </math>, where <math>q_i</math> is the <math>i</math>th column of <math>Q</math>. This is ensured by a simple random sampling and singular-value decomposition<br />
<br />
As the rows of classifier weight matrix are fixed with an equally valued <math>L_{2}</math> norm, we find it beneficial<br />
to also restrict the representation of <math>x</math> by normalizing it to reside on the <math>n</math>-dimensional sphere:<br />
<br />
<center><math><br />
\hat{x} = \frac{x}{||x||_{2}}<br />
</math></center><br />
<br />
This allows faster training and convergence, as the network does not need to account for changes in the scale of its weights. However, it has now an issue that <math>q_i · \hat{x} </math> is bounded between −1 and 1. This causes convergence issues, as the softmax function is scale sensitive, and the network is affected by the inability to re-scale its input. This issue is amended with a fixed scale <math>T</math> applied to softmax inputs <math>f(y) = softmax(\frac{1}{T}y)</math>, also known as the ''softmax temperature''. However, this introduces an additional hyper-parameter which may differ between networks and datasets. So, the authors propose to introduce a single scalar parameter <math>\alpha</math> to learn the softmax scale, effectively functioning as an inverse of the softmax temperature <math>\frac{1}{T}</math>. The normalized weights and an additional scale coefficient are also used, specially using a single scale for all entries in the weight matrix. The additional vector of bias parameters <math>b &isin; \mathbb{R}^{C}</math> is kept the same and the model is trained using the traditional negative-log-likelihood criterion. Explicitly, the classifier output is now:<br />
<br />
<center><br />
<math><br />
v_i=\frac{e^{\alpha q_i &middot; \hat{x} + b_i}}{\sum_{j}^{C} e^{\alpha q_j &middot; \hat{x} + b_j}}, i &isin; </math> { <math> {1,...,C} </math>}<br />
</center><br />
<br />
and the loss to be minimized is:<br />
<br />
<center><math><br />
L(x, t) = -\alpha q_t &middot; \frac{x}{||x||_{2}} + b_t + \text{log} (\sum_{i=1}^{C} \text{exp}((\alpha q_i &middot; \frac{x}{||x||_{2}} + b_i)))<br />
</math></center><br />
<br />
where <math>x</math> is the final representation obtained by the network for a specific sample, and <math> t &isin; </math> { <math> {1, . . . , C} </math> } is the ground-truth label for that sample. The behaviour of the parameter <math> \alpha </math> over time, which is logarithmic in nature and has the same behavior exhibited by the norm of a learned classifier, is shown in<br />
[[Media: figure1_log_behave.png| Figure 1]].<br />
<br />
<center>[[File:figure1_log_behave.png]]</center><br />
<br />
When <math> -1 \le q_i · \hat{x} \le 1 </math>, a possible cosine angle loss is <br />
<br />
<center>[[File:caloss.png]]</center><br />
<br />
But its final validation accuracy has a slight decrease, compared to original models.<br />
<br />
==Using a Hadamard Matrix==<br />
<br />
To recall, Hadamard matrix (Hedayat et al., 1978) <math> H </math> is an <math> n × n </math> matrix, where all of its entries are either +1 or −1. <br />
Example:<br />
<br />
<math>\displaystyle H_{4}={\begin{bmatrix}1&1&1&1\\1&-1&1&-1\\1&1&-1&-1\\1&-1&-1&1\end{bmatrix}}</math><br />
<br />
Furthermore, <math> H </math> is orthogonal, such that <math> HH^{T} = nI_n </math> where <math>I_n</math> is the identity matrix. Instead of using the entire Hadamard matrix <math>H</math>, a truncated version, <math> \hat{H} &isin; </math> {<math> {-1, 1}</math>}<math>^{C \times N}</math> where all <math>C</math> rows are orthogonal as the final classification layer is such that:<br />
<br />
<center><math><br />
y = \hat{H} \hat{x} + b<br />
</math></center><br />
<br />
This usage allows two main benefits:<br />
* A deterministic, low-memory and easily generated matrix that can be used for classification.<br />
* Removal of the need to perform a full matrix-matrix multiplication - as multiplying by a Hadamard matrix can be done by simple sign manipulation and addition.<br />
<br />
Here, <math>n</math> must be a multiple of 4, but it can be easily truncated to fit normally defined networks. Also, as the classifier weights are fixed to need only 1-bit precision, it is now possible to focus our attention on the features preceding it.<br />
<br />
=Experimental Results=<br />
<br />
The authors have evaluated their proposed model on the following datasets:<br />
<br />
==CIFAR-10/100==<br />
<br />
===About the Dataset===<br />
<br />
CIFAR-10 is an image classification benchmark dataset containing 50,000 training images and 10,000 test images. The images are in color and contain 32×32 pixels. There are 10 possible classes of various animals and vehicles. CIFAR-100 holds the same number of images of the same size, but contains 100 different classes.<br />
<br />
===Training Details===<br />
<br />
The authors trained a residual network ( He et al., 2016) on the CIFAR-10 dataset. The network depth was 56 and the same hyper-parameters as in the original work were used. A comparison of the two variants, i.e., the learned classifier and the proposed classifier with a fixed transformation is shown in [[Media: figure1_resnet_cifar10.png | Figure 2]].<br />
<br />
<center>[[File: figure1_resnet_cifar10.png]]</center><br />
<br />
These results demonstrate that although the training error is considerably lower for the network with learned classifier, both models achieve the same classification accuracy on the validation set. The authors' conjecture is that with the new fixed parameterization, the network can no longer increase the norm of a given sample’s representation - thus learning its label requires more effort. As this may happen for specific seen samples - it affects only training error.<br />
<br />
The authors also compared using a fixed scale variable <math>\alpha </math> at different values vs. the learned parameter. Results for <math> \alpha = </math> {0.1, 1, 10} are depicted in [[Media: figure3_alpha_resnet_cifar.png| Figure 3]] for both training and validation error and as can be seen, similar validation accuracy can be obtained using a fixed scale value (in this case <math>\alpha </math>= 1 or 10 will suffice) at the expense of another hyper-parameter to seek. In all the further experiments the scaling parameter <math> \alpha </math> was regularized with the same weight decay coefficient used on original classifier. Although learning the scale is not necessary, but it will help convergence during training.<br />
<br />
<center>[[File: figure3_alpha_resnet_cifar.png]]</center><br />
<br />
The authors then train the model on CIFAR-100 dataset. They used the DenseNet-BC model from Huang et al. (2017) with a depth of 100 layers and k = 12. The higher number of classes caused the number of parameters to grow and encompassed about 4% of the whole model. However, validation accuracy for the fixed-classifier model remained equally good as the original model, and the same training curve was observed as earlier.<br />
<br />
==IMAGENET==<br />
<br />
===About the Dataset===<br />
<br />
The Imagenet dataset introduced by Deng et al. (2009) spans over 1000 visual classes, and over 1.2 million samples. This is supposedly a more challenging dataset to work on as compared to CIFAR-10/100.<br />
<br />
===Experiment Details===<br />
<br />
The authors evaluated their fixed classifier method on Imagenet using Resnet50 by He et al. (2016) and Densenet169 model (Huang et al., 2017) as described in the original work. Using a fixed classifier removed approximately 2-million parameters were from the model, accounting for about 8% and 12 % of the model parameters respectively. The experiments revealed similar trends as observed on CIFAR-10.<br />
<br />
For a more stricter evaluation, the authors also trained a Shufflenet architecture (Zhang et al., 2017b), which was designed to be used in low memory and limited computing platforms and has parameters making up the majority of the model. They were able to reduce the parameters to 0.86 million as compared to 0.96 million parameters in the final layer of the original model. Again, the proposed modification in the original model gave similar convergence results on validation accuracy. Interestingly, this method allowed Imagenet training in an under-specified regime, where there are<br />
more training samples than the number of parameters. This is an unconventional regime for modern deep networks, which are usually over-specified to have many more parameters than training samples (Zhang et al., 2017a).<br />
<br />
The overall results of the fixed-classifier are summarized in [[Media: table1_fixed_results.png | Table 1]].<br />
<br />
<center>[[File: table1_fixed_results.png]]</center><br />
<br />
==Language Modelling==<br />
<br />
Recent works have empirically found that using the same weights for both word embedding and classifier can yield equal or better results than using a separate pair of weights. So the authors experimented with fix-classifiers on language modeling as it also requires classification of all possible tokens available in the task vocabulary. They trained a recurrent model with 2-layers of LSTM (Hochreiter & Schmidhuber, 1997) and embedding + hidden size of 512 on the WikiText2 dataset (Merity et al., 2016), using same settings as in Merity et al. (2017). WikiText2 dataset contains about 33K different words, so the number of parameters expected in the embedding and classifier layer was about 34-million. This number is about 89% of the total number of parameters used for the whole model which is 38-million. However, using a random orthogonal transform yielded poor results compared to learned embedding. This was suspected to be due to semantic relationships captured in the embedding layer of language models, which is not the case in image classification task. The intuition was further confirmed by the much better results when pre-trained embeddings using word2vec algorithm by Mikolov et al. (2013) or PMI factorization as suggested by Levy & Goldberg (2014), were used. The final result used 89% fewer parameters than a fully learned model, with only marginally worse perplexity. The authors posit that this implies a required structure in word embedding that originates from the semantic relatedness between words, and unbalanced classes. They further suggest that with more efficient ways to train word embeddings, it may be possible to mitigate the issues arising from this structure and class imbalance. <br />
<br />
<center>[[File: language.png]]</center><br />
<br />
=Discussion=<br />
<br />
==Implications and Use Cases==<br />
<br />
With the increasing number of classes in the benchmark datasets, computational demands for the final classifier will increase as well. In order to understand the problem better, the authors observe the work by Sun et al. (2017), which introduced JFT-300M - an internal Google dataset with over 18K different classes. Using a Resnet50 (He et al., 2016), with a 2048 sized representation led to a model with over 36M parameters meaning that over 60% of the model parameters resided in the final classification layer. Sun et al. (2017) also describe the difficulty in distributing so many parameters over the training servers involving a non-trivial overhead during synchronization of the model for update. The authors claim that the fixed-classifier would help considerably in this kind of scenario - where using a fixed classifier removes the need to do any gradient synchronization for the final layer. Furthermore, introduction of Hadamard matrix removes the need to save the transformation altogether, thereby, making it more efficient and allowing considerable memory and computational savings.<br />
<br />
==Possible Caveats==<br />
<br />
The good performance of fixed-classifiers relies on the ability of the preceding layers to learn separable representations. This could be affected when the ratio between learned features and number of classes is small – that is, when <math> C > N</math>. However, they tested their method in such cases and their model performed well and provided good results. <br />
<br />
Some experiments were conducted by the authors with such cases, for example Imagenet classification (C = 1000) using mobilenet-0.5 with N = 512 or using ResNet with N = 256. In both scenarios, this method converged similarly to a fully learned classifier reaching the same final validation accuracy. Although, there is no presentation of this information within the paper itself, if true, it may strengthen the finding that even in cases in which C > N, fixed classifier can provide equally good results.<br />
<br />
Another factor that can affect the performance of their model using a fixed classifier is when the classes are highly correlated. In that case, the fixed classifier actually cannot support correlated classes and thus, the network could have some difficulty to learn. For a language model, word classes tend to have highly correlated instances, which also lead to difficult learning process.<br />
<br />
Also, this proposed approach will only eliminate the computation of the classifier weights, so when the classes are fewer, the computation saving effect will not be readily apparent.<br />
<br />
==Future Work==<br />
<br />
<br />
The use of fixed classifiers might be further simplified in Binarized Neural Networks (Hubara et al., 2016a), where the activations and weights are restricted to ±1 during propagations. In that case, the norm of the last hidden layer would be constant for all samples (equal to the square root of the hidden layer width). The constant could then be absorbed into the scale constant <math>\alpha</math>, and there is no need in a per-sample normalization.<br />
<br />
Additionally, more efficient ways to learn a word embedding should also be explored where similar redundancy in classifier weights may suggest simpler forms of token representations - such as low-rank or sparse versions.<br />
<br />
A related paper was published that claims that fixing most of the parameters of the neural network achieves comparable results with learning all of them [A. Rosenfeld and J. K. Tsotsos]<br />
<br />
=Conclusion=<br />
<br />
In this work, the authors argue that the final classification layer in deep neural networks is redundant and suggest removing the parameters from the classification layer. The empirical results from experiments on the CIFAR and IMAGENET datasets suggest that such a change lead to little or almost no decline in the performance of the architecture. Furthermore, using a Hadmard matrix as classifier might lead to some computational benefits when properly implemented, and save memory otherwise spent on large amount of transformation coefficients.<br />
<br />
Another possible scope of research that could be pointed out for future could be to find new efficient methods to create pre-defined word embeddings, which require huge amount of parameters that can possibly be avoided when learning a new task. Therefore, more emphasis should be given to the representations learned by the non-linear parts of the neural networks - up to the final classifier, as it seems highly redundant.<br />
<br />
=Critique=<br />
<br />
The paper proposes an interesting idea that has a potential use case when designing memory-efficient neural networks. The experiments shown in the paper are quite rigorous and provide support to the authors' claim. However, it would have been more helpful if the authors had described a bit more about efficient implementation of the Hadamard matrix and how to scale this method for larger datasets (cases with <math> C >N</math>).<br />
<br />
Moreover, one of the main intuitions of the paper has introduced to be computational cost but it has left out to compare a fixed and learned classifier based on the computational cost and then investigate whether it worth the drop in performance or not considering the fact that not always the output can be degraded because of need for speed! At least a discussion on this issue is expected.<br />
<br />
On the other hand, the computational cost and performance change after fixation of classifier could be related to dataset and the nature and complexity of it. Mostly, having 1000 classes makes the classification more crucial than 2 classes. An evaluation of this topic is also needed.<br />
<br />
Another interesting experiment to do would be to look this technique interacts with distillation when used in the teacher or student network or both. For instance, Does fixing the features make it more difficult to place dog than on boat when classifying a cat? Do networks with fixed classifier weights make worse teachers for distillation?<br />
<br />
=References=<br />
<br />
Madhu S Advani and Andrew M Saxe. High-dimensional dynamics of generalization error in neural networks. arXiv preprint arXiv:1710.03667, 2017.<br />
<br />
Peter Bartlett, Dylan J Foster, and Matus Telgarsky. Spectrally-normalized margin bounds for neural networks. arXiv preprint arXiv:1706.08498, 2017.<br />
<br />
Jane Bromley, Isabelle Guyon, Yann LeCun, Eduard Sackinger, and Roopak Shah. Signature verification using a” siamese” time delay neural network. In Advances in Neural Information Processing Systems, pp. 737–744, 1994.<br />
<br />
Matthieu Courbariaux, Yoshua Bengio, and Jean-Pierre David. Binaryconnect: Training deep neural networks with binary weights during propagations. In Advances in Neural Information Processing Systems, pp. 3123–3131, 2015.<br />
<br />
Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In Computer Vision and Pattern Recognition, 2009. CVPR 2009. IEEE Conference on, pp. 248–255. IEEE, 2009.<br />
<br />
Suriya Gunasekar, Blake Woodworth, Srinadh Bhojanapalli, Behnam Neyshabur, and Nathan Srebro. Implicit regularization in matrix factorization. arXiv preprint arXiv:1705.09280, 2017.<br />
<br />
Song Han, Huizi Mao, and William J Dally. Deep compression: Compressing deep neural networks with pruning, trained quantization and huffman coding. arXiv preprint arXiv:1510.00149, 2015.<br />
<br />
Moritz Hardt and Tengyu Ma. Identity matters in deep learning. 2017.<br />
<br />
Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770–778, 2016.<br />
<br />
A Hedayat, WD Wallis, et al. Hadamard matrices and their applications. The Annals of Statistics, 6<br />
(6):1184–1238, 1978.<br />
<br />
Sepp Hochreiter and Jurgen Schmidhuber. Long short-term memory. ¨ Neural computation, 9(8): 1735–1780, 1997.<br />
<br />
Elad Hoffer and Nir Ailon. Deep metric learning using triplet network. In International Workshop on Similarity-Based Pattern Recognition, pp. 84–92. Springer, 2015.<br />
<br />
Elad Hoffer, Itay Hubara, and Daniel Soudry. Train longer, generalize better: closing the generalization gap in large batch training of neural networks. 2017.<br />
<br />
Andrew G Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, and Hartwig Adam. Mobilenets: Efficient convolutional neural networks for mobile vision applications. arXiv preprint arXiv:1704.04861, 2017.<br />
<br />
Gao Huang, Zhuang Liu, Laurens van der Maaten, and Kilian Q Weinberger. Densely connected convolutional networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017.<br />
<br />
Guang-Bin Huang, Qin-Yu Zhu, and Chee-Kheong Siew. Extreme learning machine: theory and applications. Neurocomputing, 70(1):489–501, 2006.<br />
<br />
Itay Hubara, Matthieu Courbariaux, Daniel Soudry, Ran El-Yaniv, and Yoshua Bengio. Binarized neural networks. In Advances in Neural Information Processing Systems 29 (NIPS’16), 2016a.<br />
<br />
Itay Hubara, Matthieu Courbariaux, Daniel Soudry, Ran El-Yaniv, and Yoshua Bengio. Quantized neural networks: Training neural networks with low precision weights and activations. arXiv preprint arXiv:1609.07061, 2016b.<br />
<br />
Hakan Inan, Khashayar Khosravi, and Richard Socher. Tying word vectors and word classifiers: A loss framework for language modeling. arXiv preprint arXiv:1611.01462, 2016.<br />
<br />
Max Jaderberg, Andrea Vedaldi, and Andrew Zisserman. Speeding up convolutional neural networks with low rank expansions. arXiv preprint arXiv:1405.3866, 2014.<br />
<br />
Alex Krizhevsky. Learning multiple layers of features from tiny images. 2009.<br />
<br />
Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks. In Advances in neural information processing systems, pp. 1097–1105, 2012.<br />
<br />
Yann LeCun, Leon Bottou, Yoshua Bengio, and Patrick Haffner. Gradient-based learning applied to ´ document recognition. Proceedings of the IEEE, 86(11):2278 2324, 1998.<br />
<br />
Omer Levy and Yoav Goldberg. Neural word embedding as implicit matrix factorization. In Advances in neural information processing systems, pp. 2177–2185, 2014.<br />
<br />
Fengfu Li, Bo Zhang, and Bin Liu. Ternary weight networks. arXiv preprint arXiv:1605.04711, 2016.<br />
<br />
Min Lin, Qiang Chen, and Shuicheng Yan. Network in network. arXiv preprint arXiv:1312.4400, 2013.<br />
<br />
Stephen Merity, Caiming Xiong, James Bradbury, and Richard Socher. Pointer sentinel mixture models. arXiv preprint arXiv:1609.07843, 2016.<br />
<br />
Stephen Merity, Nitish Shirish Keskar, and Richard Socher. Regularizing and Optimizing LSTM Language Models. arXiv preprint arXiv:1708.02182, 2017.<br />
<br />
Paulius Micikevicius, Sharan Narang, Jonah Alben, Gregory Diamos, Erich Elsen, David Garcia, Boris Ginsburg, Michael Houston, Oleksii Kuchaev, Ganesh Venkatesh, et al. Mixed precision training. arXiv preprint arXiv:1710.03740, 2017.<br />
<br />
Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S Corrado, and Jeff Dean. Distributed tations of words and phrases and their compositionality. In Advances in neural information processing systems, pp. 3111–3119, 2013.<br />
<br />
Behnam Neyshabur, Srinadh Bhojanapalli, David McAllester, and Nathan Srebro. Exploring generalization in deep learning. arXiv preprint arXiv:1706.08947, 2017.<br />
Jooyoung Park and Irwin W Sandberg. Universal approximation using radial-basis-function networks. Neural computation, 3(2):246–257, 1991.<br />
<br />
Ofir Press and Lior Wolf. Using the output embedding to improve language models. EACL 2017,<br />
pp. 157, 2017.<br />
<br />
Itay Safran and Ohad Shamir. On the quality of the initial basin in overspecified neural networks. In International Conference on Machine Learning, pp. 774–782, 2016.<br />
<br />
Tim Salimans and Diederik P Kingma. Weight normalization: A simple reparameterization to accelerate training of deep neural networks. In Advances in Neural Information Processing Systems, pp. 901–909, 2016.<br />
<br />
Florian Schroff, Dmitry Kalenichenko, and James Philbin. Facenet: A unified embedding for face recognition and clustering. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 815–823, 2015.<br />
<br />
Mahdi Soltanolkotabi, Adel Javanmard, and Jason D Lee. Theoretical insights into the optimization landscape of over-parameterized shallow neural networks. arXiv preprint arXiv:1707.04926, 2017.<br />
<br />
Daniel Soudry and Yair Carmon. No bad local minima: Data independent training error guarantees for multilayer neural networks. arXiv preprint arXiv:1605.08361, 2016.<br />
<br />
Daniel Soudry and Elad Hoffer. Exponentially vanishing sub-optimal local minima in multilayer neural networks. arXiv preprint arXiv:1702.05777, 2017.<br />
<br />
Daniel Soudry, Elad Hoffer, and Nathan Srebro. The implicit bias of gradient descent on separable data. 2018.<br />
<br />
Jost Tobias Springenberg, Alexey Dosovitskiy, Thomas Brox, and Martin Riedmiller. Striving for simplicity: The all convolutional net. arXiv preprint arXiv:1412.6806, 2014.<br />
<br />
Chen Sun, Abhinav Shrivastava, Saurabh Singh, and Abhinav Gupta. Revisiting unreasonable effectiveness of data in deep learning era. arXiv preprint arXiv:1707.02968, 2017.<br />
<br />
Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, and Andrew Rabinovich. Going deeper with convolutions. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 1–9, 2015.<br />
<br />
Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jon Shlens, and Zbigniew Wojna. Rethinking the inception architecture for computer vision. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2818–2826, 2016.<br />
<br />
Cheng Tai, Tong Xiao, Yi Zhang, Xiaogang Wang, et al. Convolutional neural networks with lowrank regularization. arXiv preprint arXiv:1511.06067, 2015.<br />
<br />
Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. 2017.<br />
Ashia C Wilson, Rebecca Roelofs, Mitchell Stern, Nathan Srebro, and Benjamin Recht. The marginal value of adaptive gradient methods in machine learning. arXiv preprint arXiv:1705.08292, 2017.<br />
<br />
Bo Xie, Yingyu Liang, and Le Song. Diversity leads to generalization in neural networks. arXiv preprint arXiv:1611.03131, 2016.<br />
<br />
Matthew D Zeiler and Rob Fergus. Visualizing and understanding convolutional networks. In European conference on computer vision, pp. 818–833. Springer, 2014. Chiyuan Zhang, Samy Bengio, Moritz Hardt, Benjamin Recht, and Oriol Vinyals. Understanding deep learning requires rethinking generalization. In ICLR, 2017a. URL https://arxiv.org/abs/1611.03530.<br />
<br />
Xiangyu Zhang, Xinyu Zhou, Mengxiao Lin, and Jian Sun. Shufflenet: An extremely efficient convolutional neural network for mobile devices. arXiv preprint arXiv:1707.01083, 2017b.<br />
<br />
Shuchang Zhou, Zekun Ni, Xinyu Zhou, He Wen, Yuxin Wu, and Yuheng Zou. Dorefa-net: Training low bitwidth convolutional neural networks with low bitwidth gradients. arXiv preprint arXiv:1606.06160, 2016.<br />
<br />
A. Rosenfeld and J. K. Tsotsos, “Intriguing properties of randomly weighted networks: Generalizing while learning next to nothing,” arXiv preprint arXiv:1802.00844, 2018.</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Unsupervised_Neural_Machine_Translation&diff=41607Unsupervised Neural Machine Translation2018-11-27T17:23:10Z<p>Gchalato: /* Methodology */</p>
<hr />
<div>This paper was published in ICLR 2018, authored by Mikel Artetxe, Gorka Labaka, Eneko Agirre, and Kyunghyun Cho. Open source implementation of this paper is available [https://github.com/artetxem/undreamt here]<br />
<br />
= Introduction =<br />
The paper presents an unsupervised Neural Machine Translation (NMT) method that uses monolingual corpora (single language texts) only. This contrasts with the usual supervised NMT approach which relies on parallel corpora (aligned text) from the source and target languages being available for training. This problem is important because parallel pairing for a majority of languages, e.g. for German-Russian, do not exist.<br />
<br />
Other authors have recently tried to address this problem using semi-supervised approaches (small set of parallel corpora). However, these methods still require a strong cross-lingual signal. The proposed method eliminates the need for cross-lingual information all together and relies solely on monolingual data. The proposed method builds upon the work done recently on unsupervised cross-lingual embeddings by Artetxe et al., 2017 and Zhang et al., 2017.<br />
<br />
The general approach of the methodology is to:<br />
<br />
# Use monolingual corpora in the source and target languages to learn single language word embeddings for both languages separately.<br />
# Align the 2 sets of word embeddings into a single cross lingual (language independent) embedding.<br />
Then iteratively perform:<br />
# Train an encoder-decoder model to reconstruct noisy versions of sentences in both source and target languages separately. The model uses a single encoder and different decoders for each language. The encoder uses cross lingual word embedding.<br />
# Tune the decoder in each language by back-translating between the source and target language.<br />
<br />
= Background =<br />
<br />
===Word Embedding Alignment===<br />
<br />
The paper uses word2vec [Mikolov, 2013] to convert each monolingual corpora to vector embeddings. They improve the continuous Skip-gram model for learning high-quality distributed vector representations that capture a large number of precise syntactic and semantic word relationships. These embeddings have been shown to contain the contextual and syntactic features independent of language, and so, in theory, there could exist a linear map that maps the embeddings from language L1 to language L2. <br />
<br />
Figure 1 shows an example of aligning the word embeddings in English and French.<br />
<br />
[[File:Figure1_lwali.png|frame|400px|center|Figure 1: the word embeddings in English and French (a & b), and (c) shows the aligned word embeddings after some linear transformation.[Gouws,2016]]]<br />
<br />
Most cross-lingual word embedding methods use bilingual signals in the form of parallel corpora. Usually, the embedding mapping methods train the embeddings in different languages using monolingual corpora, then use a linear transformation to map them into a shared space based on a bilingual dictionary.<br />
<br />
The paper uses the methodology proposed by [Artetxe, 2017] to do cross-lingual embedding aligning in an unsupervised manner and without parallel data. Without going into the details, the general approach of this paper is starting from a seed dictionary of numeral pairings (e.g. 1-1, 2-2, etc.), to iteratively learn the mapping between 2 language embeddings, while concurrently improving the dictionary with the learned mapping at each iteration. This is in contrast to earlier work which used dictionaries of a few thousand words.<br />
<br />
===Other related work and inspirations===<br />
====Statistical Decipherment for Machine Translation====<br />
There has been significant work in statistical deciphering techniques (decipherment is the discovery of the meaning of texts written in ancient or obscure languages or scripts) to develop a machine translation model from monolingual data (Ravi & Knight, 2011; Dou & Knight, 2012). These techniques treat the source language as ciphertext (encrypted or encoded information because it contains a form of the original plaintext that is unreadable by a human or computer without the proper cipher for decoding) and model the generation process of the ciphertext as a two-stage process, which includes the generation of the original English sequence and the probabilistic replacement of the words in it. This approach takes advantage of the incorporation of syntactic knowledge of the languages. The use of word embeddings has also shown improvements in statistical decipherment.<br />
<br />
====Low-Resource Neural Machine Translation====<br />
There are also proposals that use techniques other than direct parallel corpora to do NMT. Some use a third intermediate language that is well connected to the source and target languages independently. For example, if we want to translate German into Russian, we can use English as an intermediate language (German-English and then English-Russian) since there are plenty of resources to connect English and other languages. Johnson et al. (2017) show that a multilingual extension of a standard NMT architecture performs reasonably well for language pairs when no parallel data for the source and target data was used during training. Firat et al. (2016) and Chen et al. (2017) showed that the use of advanced models like teacher-student framework can be used to improve over the baseline of translating using a third intermediate language.<br />
<br />
Other works use monolingual data in combination with scarce parallel corpora. A simple but effective technique is back-translation [Sennrich et al, 2016]. First, a synthetic parallel corpus in the target language is created. Translated sentence and back translated to the source language and compared with the original sentence.<br />
<br />
The most important contribution to the problem of training an NMT model with monolingual data was from [He, 2016], which trains two agents to translate in opposite directions (e.g. French → English and English → French) and teach each other through reinforcement learning. However, this approach still required a large parallel corpus for a warm start (about 1.2 million sentences), while this paper does not use parallel data.<br />
<br />
= Methodology =<br />
<br />
The corpora data is first preprocessed in a standard way to tokenize and case the words. The authors also experimented with an alternate way of tokenizing words by using Byte-Pair Encoding (BPE) [Sennrich, 2016] (Byte pair encoding or digram coding is a simple form of data compression in which the most common pair of consecutive bytes of data is replaced with a byte that does not occur within that data). BPE has been shown to improve embeddings of rare-words. The vocabulary was limited to the most frequent 50,000 tokens (BPE tokens or words).<br />
<br />
The tokens were then converted to word embeddings using word2vec with 300 dimensions and then aligned between languages using the method proposed by [Artetxe, 2017]. The alignment method proposed by [Artetxe, 2017] is also used as a baseline to evaluate this model as discussed later in Results.<br />
<br />
The translation model uses a standard encoder-decoder model with attention. The encoder is a 2-layer bidirectional RNN, and the decoder is a 2 layer RNN. All RNNs use GRU cells with 600 hidden units. The encoder is shared by the source and target language, while the decoder is different for each language.<br />
<br />
Although the architecture uses standard models, the proposed system differs from the standard NMT through 3 aspects:<br />
<br />
#Dual structure: NMT usually are built for one direction translations English<math>\rightarrow</math>French or French<math>\rightarrow</math>English, whereas the proposed model trains both directions at the same time translating English<math>\leftrightarrow</math>French.<br />
#Shared encoder: one encoder is shared for both source and target languages in order to produce a representation in the latent space independent of language, and each decoder learns to transform the representation back to its corresponding language. <br />
#Fixed embeddings in the encoder: Most NMT systems initialize the embeddings and update them during training, whereas the proposed system trains the embeddings in the beginning and keeps these fixed throughout training, so the encoder receives language-independent representations of the words. This approach ensures that the encoder only learns how to compose the language independent representations to build representations of the larger phrases. This requires existing unsupervised methods to create embeddings using monolingual corpora as discussed in the background. In the proposed method, even though the embeddings used are cross-lingual, the vocabulary used for each language is different. This way if the same word occurs in two different languages and has a different meaning in the respective languages then each word would get a different vector in the respective languages despite being in the same vector space. <br />
<br />
[[File:Figure2_lwali.png|600px|center]]<br />
<br />
The translation model iteratively improves the encoder and decoder by performing 2 tasks: Denoising, and Back-translation.<br />
<br />
'''Note on the need for alignment:''' To train the decoders (in an admittedly “supervised” manner) we make the assumption that they decode from the same latent space. Thus, given a sentence in either language, it needs to represent it in the same latent space to allow training. However, during the back-translation training, the shared encoder stays fixed. This implies that the encoder needs to be set beforehand. For this reason, the process of embedding and alignment is needed. <br />
<br />
===Denoising===<br />
Random noise is added to the input sentences in order to allow the model to learn some structure of languages. Without noise, the model would simply learn to copy the input word by word. Noise also allows the shared encoder to compose the embeddings of both languages in a language-independent fashion, and then be decoded by the language dependent decoder.<br />
<br />
Denoising works by reconstructing a noisy version of a sentence back into the original sentence in the same language. In mathematical form, if <math>x</math> is a sentence in language L1:<br />
<br />
# Construct <math>C(x)</math>, noisy version of <math>x</math>. In the proposed model, <math>C(x)</math> is constructed by randomly swapping contiguous words. If the length of the input sequence <math>x</math> is <math>N</math>, then a total of <math>\frac{N}{2}</math> such swaps are made.<br />
# Input <math>C(x)</math> into the current iteration of the shared encoder and use decoder for L1 to get reconstructed <math>\hat{x}</math>.<br />
<br />
The training objective is to minimize the cross entropy loss between <math>{x}</math> and <math>\hat{x}</math>.<br />
<br />
In other words, the whole system is optimized to take an input sentence in a given language, encode it using the shared encoder, and reconstruct the original sentence using the decoder of that language.<br />
<br />
The proposed noise function is to perform <math>N/2</math> random swaps of words that are contiguous, where <math>N</math> is the number of words in the sentence. This noise model also helps reduce the reliance of the model on the order of words in a sentence which may be different in the source and target languages. The system will also need to correctly learn the internal structure of a language to decode the sentence into the correct order.<br />
<br />
===Back-Translation===<br />
<br />
With only denoising, the system doesn't have a goal to improve the actual translation. Back-translation works by using the decoder of the target language to create a translation, then encoding this translation and decoding again using the source decoder to reconstruct the original sentence. In mathematical form, if <math>C(x)</math> is a noisy version of sentence <math>x</math> in language L1:<br />
<br />
# Input <math>C(x)</math> into the current iteration of shared encoder and the decoder in L2 to construct translation <math>y</math> in L2,<br />
# Construct <math>C(y)</math>, noisy version of translation <math>y</math>,<br />
# Input <math>C(y)</math> into the current iteration of shared encoder and the decoder in L1 to reconstruct <math>\hat{x}</math> in L1.<br />
<br />
The training objective is to minimize the cross entropy loss between <math>{x}</math> and <math>\hat{x}</math>.<br />
<br />
This approach alleviates issues that would have resulted from the training procedure only dealing with a single language at a time. The corpus of a language is converted to a synthetic translation, and trained to predict the original sentence from this translation. <br />
<br />
Contrary to standard back-translation that uses an independent model to back-translate the entire corpus at once, the system uses mini-batches and the dual architecture to generate pseudo-translations and then train the model with the translation, improving the model iteratively as the training progresses.<br />
<br />
===Training===<br />
<br />
Training is done by alternating these 2 objectives from mini-batch to mini-batch. Each iteration would perform one mini-batch of denoising for L1, another one for L2, one mini-batch of back-translation from L1 to L2, and another one from L2 to L1. The procedure is repeated until convergence. <br />
During decoding, greedy decoding was used at training time for back-translation, but actual inference at test time was done using beam-search with a beam size of 12.<br />
<br />
Optimizer choice and other hyperparameters can be found in the paper.<br />
<br />
=Experiments and Results=<br />
<br />
The model was evaluated using the Bilingual Evaluation Understudy (BLEU) Score, which is typically used to evaluate the quality of the translation, using a reference (ground-truth) translation.<br />
<br />
The paper trained translation model under 3 different settings to compare the performance (Table 1). All training and testing data used was from a standard NMT dataset, WMT'14.<br />
<br />
[[File:Table1_lwali.png|600px|center]]<br />
<br />
The results exhibit that for the proposed system to work properly, backtranslation is necessary. The denoising technique alone is below the baseline while big improvements appear when introducing backtranslation.<br />
<br />
===Unsupervised===<br />
<br />
The model only has access to monolingual corpora, using the News Crawl corpus with articles from 2007 to 2013. The baseline for unsupervised is the method proposed by [Artetxe, 2017], which was the unsupervised word vector alignment method discussed in the Background section.<br />
<br />
The paper adds each component piece-wise when doing an evaluation to test the impact each piece has on the final score. As shown in Table 1, Unsupervised results compared to the baseline of word-by-word results are strong, with improvement between 40% to 140%. Results also show that back-translation is essential. Denoising doesn't show a big improvement however it is required for back-translation, because otherwise, back-translation would translate nonsensical sentences. The addition of backtranslation, however, does show large improvement on all tested cases.<br />
<br />
For the BPE experiment, results show it helps in some language pairs but detract in some other language pairs. This is because while BPE helped to translate some rare words, it increased the error rates in other words. It also did not perform well when translating named entities which occur infrequently.<br />
<br />
===Semi-supervised===<br />
<br />
Since there is often some small parallel data but not enough to train a Neural Machine Translation system, the authors test a semi-supervised setting with the same monolingual data from the unsupervised settings together with either 10,000 or 100,000 random sentence pairs from the News Commentary parallel corpus. The supervision is included to improve the model during the back-translation stage to directly predict sentences that are in the parallel corpus.<br />
<br />
Table 1 shows that the model can greatly benefit from the addition of a small parallel corpus to the monolingual corpora. It is surprising that semi-supervised in row 6 outperforms supervised in row 7, one possible explanation is that both the semi-supervised training set and the test set belong to the news domain, whereas the supervised training set is all domains of corpora.<br />
<br />
===Supervised===<br />
<br />
This setting provides an upper bound to the unsupervised proposed system. The data used was the combination of all parallel corpora provided at WMT 2014, which includes Europarl, Common Crawl and News Commentary for both language pairs plus the UN and the Gigaword corpus for French- English. Moreover, the authors use the same subsets of News Commentary alone to run the separate experiments in order to compare with the semi-supervised scenario.<br />
<br />
The Comparable NMT was trained using the same proposed model except it does not use monolingual corpora, and consequently, it was trained without denoising and back-translation. The proposed model under a supervised setting does much worse than the state of the NMT in row 10, which suggests that adding the additional constraints to enable unsupervised learning also limits the potential performance. To improve these results, the authors also suggest using larger models, longer training times, and incorporating several well-known NMT techniques.<br />
<br />
===Qualitative Analysis===<br />
<br />
[[File:Table2_lwali.png|600px|center]]<br />
<br />
Table 2 shows 4 examples of French to English translations, which shows that the high-quality translations are produced by the proposed system, and this system adequately models non-trivial translation relations. Example 1 and 2 show that the model is able to not only go beyond a literal word-by-word substitution but also model structural differences in the languages (ex.e, it correctly translates "l’aeroport international de Los Angeles" as "Los Angeles International Airport", and it is capable of producing high-quality translations of long and more complex sentences. However, in Example 3 and 4, the system failed to translate the months and numbers correctly and having difficulty with comprehending odd sentence structures, which means that the proposed system has limitations. Specially, the authors points that the proposed model has difficulties to preserve some concrete details from source sentences. Results also show, the proposed model's translation quality often lags behind that of a standard supervised NMT system and also there are also some cases where there are both fluency and adequacy problems that severely hinders understanding the original message from the proposed translation, suggesting that there is still room for improvement and possible future work.<br />
<br />
=Conclusions and Future Work=<br />
<br />
The paper presented an unsupervised model to perform translations with monolingual corpora by using an attention-based encoder-decoder system and training using denoise and back-translation.<br />
<br />
Although experimental results show that the proposed model is effective as an unsupervised approach, there is significant room for improvement when using the model in a supervised way, suggesting the model is limited by the architectural modifications. Some ideas for future improvement include:<br />
*Instead of using fixed cross-lingual word embeddings at the beginning which forces the encoder to learn a common representation for both languages, progressively update the weight of the embeddings as training progresses.<br />
*Decouple the shared encoder into 2 independent encoders at some point during training<br />
*Progressively reduce the noise level<br />
*Incorporate character level information into the model, which might help address some of the adequacy issues observed in our manual analysis<br />
*Use other noise/denoising techniques, and analyze their effect in relation to the typological divergences of different language pairs.<br />
<br />
= Critique =<br />
<br />
While the idea is interesting and the results are impressive for an unsupervised approach, much of the model had actually already been proposed by other papers that are referenced. The paper doesn't add a lot of new ideas but only builds on existing techniques and combines them in a different way to achieve good experimental results. The paper is not a significant algorithmic contribution. <br />
<br />
As pointed out, in order to critically analyze the effect of the algorithm, we need to formulate the algorithm in terms of mathematics.<br />
<br />
The results showed that the proposed system performed far worse than the state of the art when used in a supervised setting, which is concerning and shows that the techniques used creates a limitation and a ceiling for performance.<br />
<br />
Additionally, there was no rigorous hyperparameter exploration/optimization for the model. As a result, it is difficult to conclude whether the performance limit observed in the constrained supervised model is the absolute limit, or whether this could be overcome in both supervised/unsupervised models with the right constraints to achieve more competitive results. <br />
<br />
The best results shown are between two very closely related languages(English and French), and does much worse for English - German, even though English and German are also closely related (but less so than English and French) which suggests that the model may not be successful at translating between distant language pairs. More testing would be interesting to see.<br />
<br />
The results comparison could have shown how the semi-supervised version of the model scores compared to other semi-supervised approaches as touched on in the other works section.<br />
<br />
Their qualitative analysis just checks whether their proposed unsupervised NMT generates a sensible translation. It is limited and it needs further detailed analysis regarding the characteristics and properties of translation which is generated by unsupervised NMT.<br />
<br />
* (As pointed out by an anonymous reviewer [https://openreview.net/forum?id=Sy2ogebAW])Future work is vague: “we would like to detect and mitigate the specific causes…” “We also think that a better handling of rare words…” That’s great, but how will you do these things? Do you have specific reasons to think this, or ideas on how to approach them? Otherwise, this is just hand-waving.<br />
<br />
= References =<br />
#'''[Mikolov, 2013]''' Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S Corrado, and Jeff Dean. "Distributed representations of words and phrases and their compositionality."<br />
#'''[Artetxe, 2017]''' Mikel Artetxe, Gorka Labaka, Eneko Agirre, "Learning bilingual word embeddings with (almost) no bilingual data".<br />
#'''[Gouws,2016]''' Stephan Gouws, Yoshua Bengio, Greg Corrado, "BilBOWA: Fast Bilingual Distributed Representations without Word Alignments."<br />
#'''[He, 2016]''' Di He, Yingce Xia, Tao Qin, Liwei Wang, Nenghai Yu, Tieyan Liu, and Wei-Ying Ma. "Dual learning for machine translation."<br />
#'''[Sennrich,2016]''' Rico Sennrich and Barry Haddow and Alexandra Birch, "Neural Machine Translation of Rare Words with Subword Units."<br />
#'''[Ravi & Knight, 2011]''' Sujith Ravi and Kevin Knight, "Deciphering foreign language."<br />
#'''[Dou & Knight, 2012]''' Qing Dou and Kevin Knight, "Large scale decipherment for out-of-domain machine translation."<br />
#'''[Johnson et al. 2017]''' Melvin Johnson,et al, "Google’s multilingual neural machine translation system: Enabling zero-shot translation."<br />
#'''[Zhang et al. 2017]''' Meng Zhang, Yang Liu, Huanbo Luan, and Maosong Sun. "Adversarial training for unsupervised bilingual lexicon induction"</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Unsupervised_Neural_Machine_Translation&diff=41605Unsupervised Neural Machine Translation2018-11-27T17:22:45Z<p>Gchalato: /* Methodology */</p>
<hr />
<div>This paper was published in ICLR 2018, authored by Mikel Artetxe, Gorka Labaka, Eneko Agirre, and Kyunghyun Cho. Open source implementation of this paper is available [https://github.com/artetxem/undreamt here]<br />
<br />
= Introduction =<br />
The paper presents an unsupervised Neural Machine Translation (NMT) method that uses monolingual corpora (single language texts) only. This contrasts with the usual supervised NMT approach which relies on parallel corpora (aligned text) from the source and target languages being available for training. This problem is important because parallel pairing for a majority of languages, e.g. for German-Russian, do not exist.<br />
<br />
Other authors have recently tried to address this problem using semi-supervised approaches (small set of parallel corpora). However, these methods still require a strong cross-lingual signal. The proposed method eliminates the need for cross-lingual information all together and relies solely on monolingual data. The proposed method builds upon the work done recently on unsupervised cross-lingual embeddings by Artetxe et al., 2017 and Zhang et al., 2017.<br />
<br />
The general approach of the methodology is to:<br />
<br />
# Use monolingual corpora in the source and target languages to learn single language word embeddings for both languages separately.<br />
# Align the 2 sets of word embeddings into a single cross lingual (language independent) embedding.<br />
Then iteratively perform:<br />
# Train an encoder-decoder model to reconstruct noisy versions of sentences in both source and target languages separately. The model uses a single encoder and different decoders for each language. The encoder uses cross lingual word embedding.<br />
# Tune the decoder in each language by back-translating between the source and target language.<br />
<br />
= Background =<br />
<br />
===Word Embedding Alignment===<br />
<br />
The paper uses word2vec [Mikolov, 2013] to convert each monolingual corpora to vector embeddings. They improve the continuous Skip-gram model for learning high-quality distributed vector representations that capture a large number of precise syntactic and semantic word relationships. These embeddings have been shown to contain the contextual and syntactic features independent of language, and so, in theory, there could exist a linear map that maps the embeddings from language L1 to language L2. <br />
<br />
Figure 1 shows an example of aligning the word embeddings in English and French.<br />
<br />
[[File:Figure1_lwali.png|frame|400px|center|Figure 1: the word embeddings in English and French (a & b), and (c) shows the aligned word embeddings after some linear transformation.[Gouws,2016]]]<br />
<br />
Most cross-lingual word embedding methods use bilingual signals in the form of parallel corpora. Usually, the embedding mapping methods train the embeddings in different languages using monolingual corpora, then use a linear transformation to map them into a shared space based on a bilingual dictionary.<br />
<br />
The paper uses the methodology proposed by [Artetxe, 2017] to do cross-lingual embedding aligning in an unsupervised manner and without parallel data. Without going into the details, the general approach of this paper is starting from a seed dictionary of numeral pairings (e.g. 1-1, 2-2, etc.), to iteratively learn the mapping between 2 language embeddings, while concurrently improving the dictionary with the learned mapping at each iteration. This is in contrast to earlier work which used dictionaries of a few thousand words.<br />
<br />
===Other related work and inspirations===<br />
====Statistical Decipherment for Machine Translation====<br />
There has been significant work in statistical deciphering techniques (decipherment is the discovery of the meaning of texts written in ancient or obscure languages or scripts) to develop a machine translation model from monolingual data (Ravi & Knight, 2011; Dou & Knight, 2012). These techniques treat the source language as ciphertext (encrypted or encoded information because it contains a form of the original plaintext that is unreadable by a human or computer without the proper cipher for decoding) and model the generation process of the ciphertext as a two-stage process, which includes the generation of the original English sequence and the probabilistic replacement of the words in it. This approach takes advantage of the incorporation of syntactic knowledge of the languages. The use of word embeddings has also shown improvements in statistical decipherment.<br />
<br />
====Low-Resource Neural Machine Translation====<br />
There are also proposals that use techniques other than direct parallel corpora to do NMT. Some use a third intermediate language that is well connected to the source and target languages independently. For example, if we want to translate German into Russian, we can use English as an intermediate language (German-English and then English-Russian) since there are plenty of resources to connect English and other languages. Johnson et al. (2017) show that a multilingual extension of a standard NMT architecture performs reasonably well for language pairs when no parallel data for the source and target data was used during training. Firat et al. (2016) and Chen et al. (2017) showed that the use of advanced models like teacher-student framework can be used to improve over the baseline of translating using a third intermediate language.<br />
<br />
Other works use monolingual data in combination with scarce parallel corpora. A simple but effective technique is back-translation [Sennrich et al, 2016]. First, a synthetic parallel corpus in the target language is created. Translated sentence and back translated to the source language and compared with the original sentence.<br />
<br />
The most important contribution to the problem of training an NMT model with monolingual data was from [He, 2016], which trains two agents to translate in opposite directions (e.g. French → English and English → French) and teach each other through reinforcement learning. However, this approach still required a large parallel corpus for a warm start (about 1.2 million sentences), while this paper does not use parallel data.<br />
<br />
= Methodology =<br />
<br />
The corpora data is first preprocessed in a standard way to tokenize and case the words. The authors also experimented with an alternate way of tokenizing words by using Byte-Pair Encoding (BPE) [Sennrich, 2016] (Byte pair encoding or digram coding is a simple form of data compression in which the most common pair of consecutive bytes of data is replaced with a byte that does not occur within that data). BPE has been shown to improve embeddings of rare-words. The vocabulary was limited to the most frequent 50,000 tokens (BPE tokens or words).<br />
<br />
The tokens were then converted to word embeddings using word2vec with 300 dimensions and then aligned between languages using the method proposed by [Artetxe, 2017]. The alignment method proposed by [Artetxe, 2017] is also used as a baseline to evaluate this model as discussed later in Results.<br />
<br />
The translation model uses a standard encoder-decoder model with attention. The encoder is a 2-layer bidirectional RNN, and the decoder is a 2 layer RNN. All RNNs use GRU cells with 600 hidden units. The encoder is shared by the source and target language, while the decoder is different for each language.<br />
<br />
Although the architecture uses standard models, the proposed system differs from the standard NMT through 3 aspects:<br />
<br />
#Dual structure: NMT usually are built for one direction translations English<math>\rightarrow</math>French or French<math>\rightarrow</math>English, whereas the proposed model trains both directions at the same time translating English<math>\leftrightarrow</math>French.<br />
#Shared encoder: one encoder is shared for both source and target languages in order to produce a representation in the latent space independent of language, and each decoder learns to transform the representation back to its corresponding language. <br />
#Fixed embeddings in the encoder: Most NMT systems initialize the embeddings and update them during training, whereas the proposed system trains the embeddings in the beginning and keeps these fixed throughout training, so the encoder receives language-independent representations of the words. This approach ensures that the encoder only learns how to compose the language independent representations to build representations of the larger phrases. This requires existing unsupervised methods to create embeddings using monolingual corpora as discussed in the background. In the proposed method, even though the embeddings used are cross-lingual, the vocabulary used for each language is different. This way if the same word occurs in two different languages and has a different meaning in the respective languages then each word would get a different vector in the respective languages despite being in the same vector space. <br />
<br />
[[File:Figure2_lwali.png|600px|center]]<br />
<br />
The translation model iteratively improves the encoder and decoder by performing 2 tasks: Denoising, and Back-translation.<br />
<br />
Note on the need for alignment:<br />
<br />
To train the decoders (in an admittedly “supervised” manner) we make the assumption that they decode from the same latent space. Thus, given a sentence in either language, it needs to represent it in the same latent space to allow training. However, during the back-translation training, the shared encoder stays fixed. This implies that the encoder needs to be set beforehand. For this reason, the process of embedding and alignment is needed. <br />
<br />
===Denoising===<br />
Random noise is added to the input sentences in order to allow the model to learn some structure of languages. Without noise, the model would simply learn to copy the input word by word. Noise also allows the shared encoder to compose the embeddings of both languages in a language-independent fashion, and then be decoded by the language dependent decoder.<br />
<br />
Denoising works by reconstructing a noisy version of a sentence back into the original sentence in the same language. In mathematical form, if <math>x</math> is a sentence in language L1:<br />
<br />
# Construct <math>C(x)</math>, noisy version of <math>x</math>. In the proposed model, <math>C(x)</math> is constructed by randomly swapping contiguous words. If the length of the input sequence <math>x</math> is <math>N</math>, then a total of <math>\frac{N}{2}</math> such swaps are made.<br />
# Input <math>C(x)</math> into the current iteration of the shared encoder and use decoder for L1 to get reconstructed <math>\hat{x}</math>.<br />
<br />
The training objective is to minimize the cross entropy loss between <math>{x}</math> and <math>\hat{x}</math>.<br />
<br />
In other words, the whole system is optimized to take an input sentence in a given language, encode it using the shared encoder, and reconstruct the original sentence using the decoder of that language.<br />
<br />
The proposed noise function is to perform <math>N/2</math> random swaps of words that are contiguous, where <math>N</math> is the number of words in the sentence. This noise model also helps reduce the reliance of the model on the order of words in a sentence which may be different in the source and target languages. The system will also need to correctly learn the internal structure of a language to decode the sentence into the correct order.<br />
<br />
===Back-Translation===<br />
<br />
With only denoising, the system doesn't have a goal to improve the actual translation. Back-translation works by using the decoder of the target language to create a translation, then encoding this translation and decoding again using the source decoder to reconstruct the original sentence. In mathematical form, if <math>C(x)</math> is a noisy version of sentence <math>x</math> in language L1:<br />
<br />
# Input <math>C(x)</math> into the current iteration of shared encoder and the decoder in L2 to construct translation <math>y</math> in L2,<br />
# Construct <math>C(y)</math>, noisy version of translation <math>y</math>,<br />
# Input <math>C(y)</math> into the current iteration of shared encoder and the decoder in L1 to reconstruct <math>\hat{x}</math> in L1.<br />
<br />
The training objective is to minimize the cross entropy loss between <math>{x}</math> and <math>\hat{x}</math>.<br />
<br />
This approach alleviates issues that would have resulted from the training procedure only dealing with a single language at a time. The corpus of a language is converted to a synthetic translation, and trained to predict the original sentence from this translation. <br />
<br />
Contrary to standard back-translation that uses an independent model to back-translate the entire corpus at once, the system uses mini-batches and the dual architecture to generate pseudo-translations and then train the model with the translation, improving the model iteratively as the training progresses.<br />
<br />
===Training===<br />
<br />
Training is done by alternating these 2 objectives from mini-batch to mini-batch. Each iteration would perform one mini-batch of denoising for L1, another one for L2, one mini-batch of back-translation from L1 to L2, and another one from L2 to L1. The procedure is repeated until convergence. <br />
During decoding, greedy decoding was used at training time for back-translation, but actual inference at test time was done using beam-search with a beam size of 12.<br />
<br />
Optimizer choice and other hyperparameters can be found in the paper.<br />
<br />
=Experiments and Results=<br />
<br />
The model was evaluated using the Bilingual Evaluation Understudy (BLEU) Score, which is typically used to evaluate the quality of the translation, using a reference (ground-truth) translation.<br />
<br />
The paper trained translation model under 3 different settings to compare the performance (Table 1). All training and testing data used was from a standard NMT dataset, WMT'14.<br />
<br />
[[File:Table1_lwali.png|600px|center]]<br />
<br />
The results exhibit that for the proposed system to work properly, backtranslation is necessary. The denoising technique alone is below the baseline while big improvements appear when introducing backtranslation.<br />
<br />
===Unsupervised===<br />
<br />
The model only has access to monolingual corpora, using the News Crawl corpus with articles from 2007 to 2013. The baseline for unsupervised is the method proposed by [Artetxe, 2017], which was the unsupervised word vector alignment method discussed in the Background section.<br />
<br />
The paper adds each component piece-wise when doing an evaluation to test the impact each piece has on the final score. As shown in Table 1, Unsupervised results compared to the baseline of word-by-word results are strong, with improvement between 40% to 140%. Results also show that back-translation is essential. Denoising doesn't show a big improvement however it is required for back-translation, because otherwise, back-translation would translate nonsensical sentences. The addition of backtranslation, however, does show large improvement on all tested cases.<br />
<br />
For the BPE experiment, results show it helps in some language pairs but detract in some other language pairs. This is because while BPE helped to translate some rare words, it increased the error rates in other words. It also did not perform well when translating named entities which occur infrequently.<br />
<br />
===Semi-supervised===<br />
<br />
Since there is often some small parallel data but not enough to train a Neural Machine Translation system, the authors test a semi-supervised setting with the same monolingual data from the unsupervised settings together with either 10,000 or 100,000 random sentence pairs from the News Commentary parallel corpus. The supervision is included to improve the model during the back-translation stage to directly predict sentences that are in the parallel corpus.<br />
<br />
Table 1 shows that the model can greatly benefit from the addition of a small parallel corpus to the monolingual corpora. It is surprising that semi-supervised in row 6 outperforms supervised in row 7, one possible explanation is that both the semi-supervised training set and the test set belong to the news domain, whereas the supervised training set is all domains of corpora.<br />
<br />
===Supervised===<br />
<br />
This setting provides an upper bound to the unsupervised proposed system. The data used was the combination of all parallel corpora provided at WMT 2014, which includes Europarl, Common Crawl and News Commentary for both language pairs plus the UN and the Gigaword corpus for French- English. Moreover, the authors use the same subsets of News Commentary alone to run the separate experiments in order to compare with the semi-supervised scenario.<br />
<br />
The Comparable NMT was trained using the same proposed model except it does not use monolingual corpora, and consequently, it was trained without denoising and back-translation. The proposed model under a supervised setting does much worse than the state of the NMT in row 10, which suggests that adding the additional constraints to enable unsupervised learning also limits the potential performance. To improve these results, the authors also suggest using larger models, longer training times, and incorporating several well-known NMT techniques.<br />
<br />
===Qualitative Analysis===<br />
<br />
[[File:Table2_lwali.png|600px|center]]<br />
<br />
Table 2 shows 4 examples of French to English translations, which shows that the high-quality translations are produced by the proposed system, and this system adequately models non-trivial translation relations. Example 1 and 2 show that the model is able to not only go beyond a literal word-by-word substitution but also model structural differences in the languages (ex.e, it correctly translates "l’aeroport international de Los Angeles" as "Los Angeles International Airport", and it is capable of producing high-quality translations of long and more complex sentences. However, in Example 3 and 4, the system failed to translate the months and numbers correctly and having difficulty with comprehending odd sentence structures, which means that the proposed system has limitations. Specially, the authors points that the proposed model has difficulties to preserve some concrete details from source sentences. Results also show, the proposed model's translation quality often lags behind that of a standard supervised NMT system and also there are also some cases where there are both fluency and adequacy problems that severely hinders understanding the original message from the proposed translation, suggesting that there is still room for improvement and possible future work.<br />
<br />
=Conclusions and Future Work=<br />
<br />
The paper presented an unsupervised model to perform translations with monolingual corpora by using an attention-based encoder-decoder system and training using denoise and back-translation.<br />
<br />
Although experimental results show that the proposed model is effective as an unsupervised approach, there is significant room for improvement when using the model in a supervised way, suggesting the model is limited by the architectural modifications. Some ideas for future improvement include:<br />
*Instead of using fixed cross-lingual word embeddings at the beginning which forces the encoder to learn a common representation for both languages, progressively update the weight of the embeddings as training progresses.<br />
*Decouple the shared encoder into 2 independent encoders at some point during training<br />
*Progressively reduce the noise level<br />
*Incorporate character level information into the model, which might help address some of the adequacy issues observed in our manual analysis<br />
*Use other noise/denoising techniques, and analyze their effect in relation to the typological divergences of different language pairs.<br />
<br />
= Critique =<br />
<br />
While the idea is interesting and the results are impressive for an unsupervised approach, much of the model had actually already been proposed by other papers that are referenced. The paper doesn't add a lot of new ideas but only builds on existing techniques and combines them in a different way to achieve good experimental results. The paper is not a significant algorithmic contribution. <br />
<br />
As pointed out, in order to critically analyze the effect of the algorithm, we need to formulate the algorithm in terms of mathematics.<br />
<br />
The results showed that the proposed system performed far worse than the state of the art when used in a supervised setting, which is concerning and shows that the techniques used creates a limitation and a ceiling for performance.<br />
<br />
Additionally, there was no rigorous hyperparameter exploration/optimization for the model. As a result, it is difficult to conclude whether the performance limit observed in the constrained supervised model is the absolute limit, or whether this could be overcome in both supervised/unsupervised models with the right constraints to achieve more competitive results. <br />
<br />
The best results shown are between two very closely related languages(English and French), and does much worse for English - German, even though English and German are also closely related (but less so than English and French) which suggests that the model may not be successful at translating between distant language pairs. More testing would be interesting to see.<br />
<br />
The results comparison could have shown how the semi-supervised version of the model scores compared to other semi-supervised approaches as touched on in the other works section.<br />
<br />
Their qualitative analysis just checks whether their proposed unsupervised NMT generates a sensible translation. It is limited and it needs further detailed analysis regarding the characteristics and properties of translation which is generated by unsupervised NMT.<br />
<br />
* (As pointed out by an anonymous reviewer [https://openreview.net/forum?id=Sy2ogebAW])Future work is vague: “we would like to detect and mitigate the specific causes…” “We also think that a better handling of rare words…” That’s great, but how will you do these things? Do you have specific reasons to think this, or ideas on how to approach them? Otherwise, this is just hand-waving.<br />
<br />
= References =<br />
#'''[Mikolov, 2013]''' Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S Corrado, and Jeff Dean. "Distributed representations of words and phrases and their compositionality."<br />
#'''[Artetxe, 2017]''' Mikel Artetxe, Gorka Labaka, Eneko Agirre, "Learning bilingual word embeddings with (almost) no bilingual data".<br />
#'''[Gouws,2016]''' Stephan Gouws, Yoshua Bengio, Greg Corrado, "BilBOWA: Fast Bilingual Distributed Representations without Word Alignments."<br />
#'''[He, 2016]''' Di He, Yingce Xia, Tao Qin, Liwei Wang, Nenghai Yu, Tieyan Liu, and Wei-Ying Ma. "Dual learning for machine translation."<br />
#'''[Sennrich,2016]''' Rico Sennrich and Barry Haddow and Alexandra Birch, "Neural Machine Translation of Rare Words with Subword Units."<br />
#'''[Ravi & Knight, 2011]''' Sujith Ravi and Kevin Knight, "Deciphering foreign language."<br />
#'''[Dou & Knight, 2012]''' Qing Dou and Kevin Knight, "Large scale decipherment for out-of-domain machine translation."<br />
#'''[Johnson et al. 2017]''' Melvin Johnson,et al, "Google’s multilingual neural machine translation system: Enabling zero-shot translation."<br />
#'''[Zhang et al. 2017]''' Meng Zhang, Yang Liu, Huanbo Luan, and Maosong Sun. "Adversarial training for unsupervised bilingual lexicon induction"</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Unsupervised_Neural_Machine_Translation&diff=41604Unsupervised Neural Machine Translation2018-11-27T17:22:30Z<p>Gchalato: /* Methodology */</p>
<hr />
<div>This paper was published in ICLR 2018, authored by Mikel Artetxe, Gorka Labaka, Eneko Agirre, and Kyunghyun Cho. Open source implementation of this paper is available [https://github.com/artetxem/undreamt here]<br />
<br />
= Introduction =<br />
The paper presents an unsupervised Neural Machine Translation (NMT) method that uses monolingual corpora (single language texts) only. This contrasts with the usual supervised NMT approach which relies on parallel corpora (aligned text) from the source and target languages being available for training. This problem is important because parallel pairing for a majority of languages, e.g. for German-Russian, do not exist.<br />
<br />
Other authors have recently tried to address this problem using semi-supervised approaches (small set of parallel corpora). However, these methods still require a strong cross-lingual signal. The proposed method eliminates the need for cross-lingual information all together and relies solely on monolingual data. The proposed method builds upon the work done recently on unsupervised cross-lingual embeddings by Artetxe et al., 2017 and Zhang et al., 2017.<br />
<br />
The general approach of the methodology is to:<br />
<br />
# Use monolingual corpora in the source and target languages to learn single language word embeddings for both languages separately.<br />
# Align the 2 sets of word embeddings into a single cross lingual (language independent) embedding.<br />
Then iteratively perform:<br />
# Train an encoder-decoder model to reconstruct noisy versions of sentences in both source and target languages separately. The model uses a single encoder and different decoders for each language. The encoder uses cross lingual word embedding.<br />
# Tune the decoder in each language by back-translating between the source and target language.<br />
<br />
= Background =<br />
<br />
===Word Embedding Alignment===<br />
<br />
The paper uses word2vec [Mikolov, 2013] to convert each monolingual corpora to vector embeddings. They improve the continuous Skip-gram model for learning high-quality distributed vector representations that capture a large number of precise syntactic and semantic word relationships. These embeddings have been shown to contain the contextual and syntactic features independent of language, and so, in theory, there could exist a linear map that maps the embeddings from language L1 to language L2. <br />
<br />
Figure 1 shows an example of aligning the word embeddings in English and French.<br />
<br />
[[File:Figure1_lwali.png|frame|400px|center|Figure 1: the word embeddings in English and French (a & b), and (c) shows the aligned word embeddings after some linear transformation.[Gouws,2016]]]<br />
<br />
Most cross-lingual word embedding methods use bilingual signals in the form of parallel corpora. Usually, the embedding mapping methods train the embeddings in different languages using monolingual corpora, then use a linear transformation to map them into a shared space based on a bilingual dictionary.<br />
<br />
The paper uses the methodology proposed by [Artetxe, 2017] to do cross-lingual embedding aligning in an unsupervised manner and without parallel data. Without going into the details, the general approach of this paper is starting from a seed dictionary of numeral pairings (e.g. 1-1, 2-2, etc.), to iteratively learn the mapping between 2 language embeddings, while concurrently improving the dictionary with the learned mapping at each iteration. This is in contrast to earlier work which used dictionaries of a few thousand words.<br />
<br />
===Other related work and inspirations===<br />
====Statistical Decipherment for Machine Translation====<br />
There has been significant work in statistical deciphering techniques (decipherment is the discovery of the meaning of texts written in ancient or obscure languages or scripts) to develop a machine translation model from monolingual data (Ravi & Knight, 2011; Dou & Knight, 2012). These techniques treat the source language as ciphertext (encrypted or encoded information because it contains a form of the original plaintext that is unreadable by a human or computer without the proper cipher for decoding) and model the generation process of the ciphertext as a two-stage process, which includes the generation of the original English sequence and the probabilistic replacement of the words in it. This approach takes advantage of the incorporation of syntactic knowledge of the languages. The use of word embeddings has also shown improvements in statistical decipherment.<br />
<br />
====Low-Resource Neural Machine Translation====<br />
There are also proposals that use techniques other than direct parallel corpora to do NMT. Some use a third intermediate language that is well connected to the source and target languages independently. For example, if we want to translate German into Russian, we can use English as an intermediate language (German-English and then English-Russian) since there are plenty of resources to connect English and other languages. Johnson et al. (2017) show that a multilingual extension of a standard NMT architecture performs reasonably well for language pairs when no parallel data for the source and target data was used during training. Firat et al. (2016) and Chen et al. (2017) showed that the use of advanced models like teacher-student framework can be used to improve over the baseline of translating using a third intermediate language.<br />
<br />
Other works use monolingual data in combination with scarce parallel corpora. A simple but effective technique is back-translation [Sennrich et al, 2016]. First, a synthetic parallel corpus in the target language is created. Translated sentence and back translated to the source language and compared with the original sentence.<br />
<br />
The most important contribution to the problem of training an NMT model with monolingual data was from [He, 2016], which trains two agents to translate in opposite directions (e.g. French → English and English → French) and teach each other through reinforcement learning. However, this approach still required a large parallel corpus for a warm start (about 1.2 million sentences), while this paper does not use parallel data.<br />
<br />
= Methodology =<br />
<br />
The corpora data is first preprocessed in a standard way to tokenize and case the words. The authors also experimented with an alternate way of tokenizing words by using Byte-Pair Encoding (BPE) [Sennrich, 2016] (Byte pair encoding or digram coding is a simple form of data compression in which the most common pair of consecutive bytes of data is replaced with a byte that does not occur within that data). BPE has been shown to improve embeddings of rare-words. The vocabulary was limited to the most frequent 50,000 tokens (BPE tokens or words).<br />
<br />
The tokens were then converted to word embeddings using word2vec with 300 dimensions and then aligned between languages using the method proposed by [Artetxe, 2017]. The alignment method proposed by [Artetxe, 2017] is also used as a baseline to evaluate this model as discussed later in Results.<br />
<br />
The translation model uses a standard encoder-decoder model with attention. The encoder is a 2-layer bidirectional RNN, and the decoder is a 2 layer RNN. All RNNs use GRU cells with 600 hidden units. The encoder is shared by the source and target language, while the decoder is different for each language.<br />
<br />
Although the architecture uses standard models, the proposed system differs from the standard NMT through 3 aspects:<br />
<br />
#Dual structure: NMT usually are built for one direction translations English<math>\rightarrow</math>French or French<math>\rightarrow</math>English, whereas the proposed model trains both directions at the same time translating English<math>\leftrightarrow</math>French.<br />
#Shared encoder: one encoder is shared for both source and target languages in order to produce a representation in the latent space independent of language, and each decoder learns to transform the representation back to its corresponding language. <br />
#Fixed embeddings in the encoder: Most NMT systems initialize the embeddings and update them during training, whereas the proposed system trains the embeddings in the beginning and keeps these fixed throughout training, so the encoder receives language-independent representations of the words. This approach ensures that the encoder only learns how to compose the language independent representations to build representations of the larger phrases. This requires existing unsupervised methods to create embeddings using monolingual corpora as discussed in the background. In the proposed method, even though the embeddings used are cross-lingual, the vocabulary used for each language is different. This way if the same word occurs in two different languages and has a different meaning in the respective languages then each word would get a different vector in the respective languages despite being in the same vector space. <br />
<br />
[[File:Figure2_lwali.png|600px|center]]<br />
<br />
The translation model iteratively improves the encoder and decoder by performing 2 tasks: Denoising, and Back-translation.<br />
<br />
Note on the need for alignment:<br />
To train the decoders (in an admittedly “supervised” manner) we make the assumption that they decode from the same latent space. Thus, given a sentence in either language, it needs to represent it in the same latent space to allow training. However, during the back-translation training, the shared encoder stays fixed. This implies that the encoder needs to be set beforehand. For this reason, the process of embedding and alignment is needed. <br />
<br />
===Denoising===<br />
Random noise is added to the input sentences in order to allow the model to learn some structure of languages. Without noise, the model would simply learn to copy the input word by word. Noise also allows the shared encoder to compose the embeddings of both languages in a language-independent fashion, and then be decoded by the language dependent decoder.<br />
<br />
Denoising works by reconstructing a noisy version of a sentence back into the original sentence in the same language. In mathematical form, if <math>x</math> is a sentence in language L1:<br />
<br />
# Construct <math>C(x)</math>, noisy version of <math>x</math>. In the proposed model, <math>C(x)</math> is constructed by randomly swapping contiguous words. If the length of the input sequence <math>x</math> is <math>N</math>, then a total of <math>\frac{N}{2}</math> such swaps are made.<br />
# Input <math>C(x)</math> into the current iteration of the shared encoder and use decoder for L1 to get reconstructed <math>\hat{x}</math>.<br />
<br />
The training objective is to minimize the cross entropy loss between <math>{x}</math> and <math>\hat{x}</math>.<br />
<br />
In other words, the whole system is optimized to take an input sentence in a given language, encode it using the shared encoder, and reconstruct the original sentence using the decoder of that language.<br />
<br />
The proposed noise function is to perform <math>N/2</math> random swaps of words that are contiguous, where <math>N</math> is the number of words in the sentence. This noise model also helps reduce the reliance of the model on the order of words in a sentence which may be different in the source and target languages. The system will also need to correctly learn the internal structure of a language to decode the sentence into the correct order.<br />
<br />
===Back-Translation===<br />
<br />
With only denoising, the system doesn't have a goal to improve the actual translation. Back-translation works by using the decoder of the target language to create a translation, then encoding this translation and decoding again using the source decoder to reconstruct the original sentence. In mathematical form, if <math>C(x)</math> is a noisy version of sentence <math>x</math> in language L1:<br />
<br />
# Input <math>C(x)</math> into the current iteration of shared encoder and the decoder in L2 to construct translation <math>y</math> in L2,<br />
# Construct <math>C(y)</math>, noisy version of translation <math>y</math>,<br />
# Input <math>C(y)</math> into the current iteration of shared encoder and the decoder in L1 to reconstruct <math>\hat{x}</math> in L1.<br />
<br />
The training objective is to minimize the cross entropy loss between <math>{x}</math> and <math>\hat{x}</math>.<br />
<br />
This approach alleviates issues that would have resulted from the training procedure only dealing with a single language at a time. The corpus of a language is converted to a synthetic translation, and trained to predict the original sentence from this translation. <br />
<br />
Contrary to standard back-translation that uses an independent model to back-translate the entire corpus at once, the system uses mini-batches and the dual architecture to generate pseudo-translations and then train the model with the translation, improving the model iteratively as the training progresses.<br />
<br />
===Training===<br />
<br />
Training is done by alternating these 2 objectives from mini-batch to mini-batch. Each iteration would perform one mini-batch of denoising for L1, another one for L2, one mini-batch of back-translation from L1 to L2, and another one from L2 to L1. The procedure is repeated until convergence. <br />
During decoding, greedy decoding was used at training time for back-translation, but actual inference at test time was done using beam-search with a beam size of 12.<br />
<br />
Optimizer choice and other hyperparameters can be found in the paper.<br />
<br />
=Experiments and Results=<br />
<br />
The model was evaluated using the Bilingual Evaluation Understudy (BLEU) Score, which is typically used to evaluate the quality of the translation, using a reference (ground-truth) translation.<br />
<br />
The paper trained translation model under 3 different settings to compare the performance (Table 1). All training and testing data used was from a standard NMT dataset, WMT'14.<br />
<br />
[[File:Table1_lwali.png|600px|center]]<br />
<br />
The results exhibit that for the proposed system to work properly, backtranslation is necessary. The denoising technique alone is below the baseline while big improvements appear when introducing backtranslation.<br />
<br />
===Unsupervised===<br />
<br />
The model only has access to monolingual corpora, using the News Crawl corpus with articles from 2007 to 2013. The baseline for unsupervised is the method proposed by [Artetxe, 2017], which was the unsupervised word vector alignment method discussed in the Background section.<br />
<br />
The paper adds each component piece-wise when doing an evaluation to test the impact each piece has on the final score. As shown in Table 1, Unsupervised results compared to the baseline of word-by-word results are strong, with improvement between 40% to 140%. Results also show that back-translation is essential. Denoising doesn't show a big improvement however it is required for back-translation, because otherwise, back-translation would translate nonsensical sentences. The addition of backtranslation, however, does show large improvement on all tested cases.<br />
<br />
For the BPE experiment, results show it helps in some language pairs but detract in some other language pairs. This is because while BPE helped to translate some rare words, it increased the error rates in other words. It also did not perform well when translating named entities which occur infrequently.<br />
<br />
===Semi-supervised===<br />
<br />
Since there is often some small parallel data but not enough to train a Neural Machine Translation system, the authors test a semi-supervised setting with the same monolingual data from the unsupervised settings together with either 10,000 or 100,000 random sentence pairs from the News Commentary parallel corpus. The supervision is included to improve the model during the back-translation stage to directly predict sentences that are in the parallel corpus.<br />
<br />
Table 1 shows that the model can greatly benefit from the addition of a small parallel corpus to the monolingual corpora. It is surprising that semi-supervised in row 6 outperforms supervised in row 7, one possible explanation is that both the semi-supervised training set and the test set belong to the news domain, whereas the supervised training set is all domains of corpora.<br />
<br />
===Supervised===<br />
<br />
This setting provides an upper bound to the unsupervised proposed system. The data used was the combination of all parallel corpora provided at WMT 2014, which includes Europarl, Common Crawl and News Commentary for both language pairs plus the UN and the Gigaword corpus for French- English. Moreover, the authors use the same subsets of News Commentary alone to run the separate experiments in order to compare with the semi-supervised scenario.<br />
<br />
The Comparable NMT was trained using the same proposed model except it does not use monolingual corpora, and consequently, it was trained without denoising and back-translation. The proposed model under a supervised setting does much worse than the state of the NMT in row 10, which suggests that adding the additional constraints to enable unsupervised learning also limits the potential performance. To improve these results, the authors also suggest using larger models, longer training times, and incorporating several well-known NMT techniques.<br />
<br />
===Qualitative Analysis===<br />
<br />
[[File:Table2_lwali.png|600px|center]]<br />
<br />
Table 2 shows 4 examples of French to English translations, which shows that the high-quality translations are produced by the proposed system, and this system adequately models non-trivial translation relations. Example 1 and 2 show that the model is able to not only go beyond a literal word-by-word substitution but also model structural differences in the languages (ex.e, it correctly translates "l’aeroport international de Los Angeles" as "Los Angeles International Airport", and it is capable of producing high-quality translations of long and more complex sentences. However, in Example 3 and 4, the system failed to translate the months and numbers correctly and having difficulty with comprehending odd sentence structures, which means that the proposed system has limitations. Specially, the authors points that the proposed model has difficulties to preserve some concrete details from source sentences. Results also show, the proposed model's translation quality often lags behind that of a standard supervised NMT system and also there are also some cases where there are both fluency and adequacy problems that severely hinders understanding the original message from the proposed translation, suggesting that there is still room for improvement and possible future work.<br />
<br />
=Conclusions and Future Work=<br />
<br />
The paper presented an unsupervised model to perform translations with monolingual corpora by using an attention-based encoder-decoder system and training using denoise and back-translation.<br />
<br />
Although experimental results show that the proposed model is effective as an unsupervised approach, there is significant room for improvement when using the model in a supervised way, suggesting the model is limited by the architectural modifications. Some ideas for future improvement include:<br />
*Instead of using fixed cross-lingual word embeddings at the beginning which forces the encoder to learn a common representation for both languages, progressively update the weight of the embeddings as training progresses.<br />
*Decouple the shared encoder into 2 independent encoders at some point during training<br />
*Progressively reduce the noise level<br />
*Incorporate character level information into the model, which might help address some of the adequacy issues observed in our manual analysis<br />
*Use other noise/denoising techniques, and analyze their effect in relation to the typological divergences of different language pairs.<br />
<br />
= Critique =<br />
<br />
While the idea is interesting and the results are impressive for an unsupervised approach, much of the model had actually already been proposed by other papers that are referenced. The paper doesn't add a lot of new ideas but only builds on existing techniques and combines them in a different way to achieve good experimental results. The paper is not a significant algorithmic contribution. <br />
<br />
As pointed out, in order to critically analyze the effect of the algorithm, we need to formulate the algorithm in terms of mathematics.<br />
<br />
The results showed that the proposed system performed far worse than the state of the art when used in a supervised setting, which is concerning and shows that the techniques used creates a limitation and a ceiling for performance.<br />
<br />
Additionally, there was no rigorous hyperparameter exploration/optimization for the model. As a result, it is difficult to conclude whether the performance limit observed in the constrained supervised model is the absolute limit, or whether this could be overcome in both supervised/unsupervised models with the right constraints to achieve more competitive results. <br />
<br />
The best results shown are between two very closely related languages(English and French), and does much worse for English - German, even though English and German are also closely related (but less so than English and French) which suggests that the model may not be successful at translating between distant language pairs. More testing would be interesting to see.<br />
<br />
The results comparison could have shown how the semi-supervised version of the model scores compared to other semi-supervised approaches as touched on in the other works section.<br />
<br />
Their qualitative analysis just checks whether their proposed unsupervised NMT generates a sensible translation. It is limited and it needs further detailed analysis regarding the characteristics and properties of translation which is generated by unsupervised NMT.<br />
<br />
* (As pointed out by an anonymous reviewer [https://openreview.net/forum?id=Sy2ogebAW])Future work is vague: “we would like to detect and mitigate the specific causes…” “We also think that a better handling of rare words…” That’s great, but how will you do these things? Do you have specific reasons to think this, or ideas on how to approach them? Otherwise, this is just hand-waving.<br />
<br />
= References =<br />
#'''[Mikolov, 2013]''' Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S Corrado, and Jeff Dean. "Distributed representations of words and phrases and their compositionality."<br />
#'''[Artetxe, 2017]''' Mikel Artetxe, Gorka Labaka, Eneko Agirre, "Learning bilingual word embeddings with (almost) no bilingual data".<br />
#'''[Gouws,2016]''' Stephan Gouws, Yoshua Bengio, Greg Corrado, "BilBOWA: Fast Bilingual Distributed Representations without Word Alignments."<br />
#'''[He, 2016]''' Di He, Yingce Xia, Tao Qin, Liwei Wang, Nenghai Yu, Tieyan Liu, and Wei-Ying Ma. "Dual learning for machine translation."<br />
#'''[Sennrich,2016]''' Rico Sennrich and Barry Haddow and Alexandra Birch, "Neural Machine Translation of Rare Words with Subword Units."<br />
#'''[Ravi & Knight, 2011]''' Sujith Ravi and Kevin Knight, "Deciphering foreign language."<br />
#'''[Dou & Knight, 2012]''' Qing Dou and Kevin Knight, "Large scale decipherment for out-of-domain machine translation."<br />
#'''[Johnson et al. 2017]''' Melvin Johnson,et al, "Google’s multilingual neural machine translation system: Enabling zero-shot translation."<br />
#'''[Zhang et al. 2017]''' Meng Zhang, Yang Liu, Huanbo Luan, and Maosong Sun. "Adversarial training for unsupervised bilingual lexicon induction"</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Mapping_Images_to_Scene_Graphs_with_Permutation-Invariant_Structured_Prediction&diff=41062Mapping Images to Scene Graphs with Permutation-Invariant Structured Prediction2018-11-23T02:19:52Z<p>Gchalato: /* Permutation-Invariant Structured prediction */</p>
<hr />
<div>The paper ''Mapping Images to Scene Graphs with Permutation-Invariant Structured Prediction'' was written by Roei Herzig* from Tel Aviv University, Moshiko Raboh* from Tel Aviv University, Gal Chechik from Google Brain, Bar-Ilan University, Jonathan Berant from Tel Aviv University, and Amir Globerson from Tel Aviv University. This paper is part of the NIPS 2018 conference to be hosted in December 2018 at Montréal, Canada. This paper summary is based on version 3 of the pre-print (as of May 2018) obtained from [https://arxiv.org/pdf/1802.05451v3.pdf arXiv] <br />
<br />
(*) Equal contribution<br />
<br />
=Motivation=<br />
In the field of artificial intelligence, a major goal is to enable machines to understand complex images, such as the underlying relationships between objects that exist in each scene. Although there are models today that capture both complex labels and interactions between labels, there is a disconnect for what guidelines should be used when leveraging deep learning. This paper introduces a design principle for such models that stem from the concept of permutation invariance and proves state of the art performance on models that follow this principle.<br />
<br />
The primary contributions that this paper makes include:<br />
# Deriving sufficient and necessary conditions for respecting graph-permutation invariance in deep structured prediction architectures<br />
# Empirically proving the benefit of graph-permutation invariance<br />
# Developing a state-of-the-art model for scene graph predictions over a large set of complex visual scenes<br />
<br />
=Introduction=<br />
In order to make a machine to interpret complex visual scenes, it must recognize and understand both objects and relationships between the objects in the scene. A '''scene graph''' is a representation of the set of objects and relations that exist in the scene, where objects are represented as nodes, relations are represented as edges connecting the different nodes. Hence, the prediction of the scene graph is analogous to inferring the joint set of objects and relations of a visual scene.<br />
<br />
[[File:scene_graph_example.png|600px|center]]<br />
<br />
Given that objects in scenes are interdependent on each other, joint prediction of the objects and relations is necessary. The field of structured prediction, which involves the general problem of inferring multiple inter-dependent labels, is of interest for this problem.<br />
<br />
In structured prediction models, a score function <math>s(x, y)</math> is defined to evaluate the compatibility between label <math>y</math> and input <math>x</math>. For instance, when interpreting the scene of an image, <math>x</math> refers to the image itself, and <math>y</math> refers to a complex label, which contains both the objects and the relations between objects. As with most other inference methods, the goal is to find the label <math>y^*</math> such that <math>s(x,y)</math> is maximized, <math> y^*=argmax_y s(x,y)</math>. However, the major concern is that the space for possible label assignments grows exponentially with respect to input size. For example, although an image may seem very simple, the corpus containing possible labels for objects may be very large, rendering it difficult to optimize the scoring function. <br />
<br />
The paper presents an alternative approach, for which input <math>x</math> is mapped to structured output <math>y</math> using a "black box" neural network, omitting the definition of a score function. The main concern for this approach is the determination of the network architecture.<br />
<br />
The model is evaluated by firstly demonstrating the importance of permutation invariance on a synthetic data set. The approach laid out by the authors is then shown to respect permutation invariance, and results are compared to a competitive benchmark. This method achieves state-of-the-art results.<br />
<br />
=Structured prediction=<br />
This paper further considers structured predictions using score-based methods. For structured predictions that follow a score-based approach, a score function <math>s(x, y)</math> is used to measure how compatible label <math>y</math> is for input <math>x</math> and is also used to infer a label by maximizing <math>s(x, y)</math>. To optimize the score function, previous works have decomposed <math>s(x,y) = \sum_i f_i(x,y)</math> in order to facilitate efficient optimization which is done by optimizing the local score function, <math>\max_y f_i(x,y)</math>, with a small subset of the <math>y</math> variables.<br />
<br />
Recently, modeling the <math>f_i </math> functions as deep networks is a new interest. In such area of structured predictions, the most commonly-used score functions include the singleton score function <math>f_i(y_i, x)</math> and pairwise score function <math>f_{ij} (y_i, y_j, x)</math>. Previous works explored a two-stage architectures (learn local scores independently of the structured prediction goal), end-to-end architectures (to include the inference algorithm within the computation graph), and modelling global factors. <br />
<br />
==Advantages of using score-based methods==<br />
# Allow for intuitive specification of local dependencies between labels, and how they map to global dependencies<br />
# Linear score functions offer natural convex surrogates<br />
# Inference in large label space is sometimes possible via exact algorithms or empirically accurate approximations<br />
<br />
The concern for modelling score functions using deep networks is that learning may no longer be convex. Hence, the paper presents properties for how deep networks can be used for structured predictions by considering architectures that do not require explicit maximization of a score function.<br />
<br />
=Background, Notations, and Definitions=<br />
We denote <math>y</math> as a structured label where <math>y = [y_1, \dots, y_n]</math><br />
<br />
'''Score functions:''' for score-based methods, the score is defined as either the sum of a set of singleton scores <math>f_i = f_i(y_i, x)</math> or the sum of pairwise scores <math>f_{ij} = f_{ij}(y_i, y_j, x)</math>.<br />
<br />
Let <math>s(x,y)</math> be the score of a score-based method. Then:<br />
<br />
<div align="center"><br />
<math>s(x,y) = \begin{cases}<br />
\sum_i f_i ~ \text{if we have a set of singleton scores}\\<br />
\sum_{ij} f_{ij} ~ \text{if we have a set of pairwise scores } \\<br />
\end{cases}</math><br />
</div><br />
<br />
'''Inference algorithm:''' an inference algorithm takes input set of local scores (either <math>f_i</math> or <math>f_{ij}</math>) and outputs an assignment of labels <math>y_1, \dots, y_n</math> that maximizes score function <math>s(x,y)</math><br />
<br />
'''Graph labeling function:''' a graph labeling function <math>\mathcal{F} : (V,E) \rightarrow Y</math> is a function that takes input of: an ordered set of node features <math>V = [z_1, \dots, z_n]</math> and an ordered set of edge features <math>E = [z_{1,2},\dots,z_{i,j},\dots,z_{n,n-1}]</math> to output set of node labels <math>\mathbf{y} = [y_1, \dots, y_n]</math>. For instance, <math>z_i</math> can be set equal to <math>f_i</math> and <math>z_{ij}</math> can be set equal to <math>f_{ij}</math>.<br />
<br />
For convenience, the joint set of nodes and edges will be denoted as <math>\mathbf{z}</math> to be a size <math>n^2</math> vector (<math>n</math> nodes and <math>n(n-1)</math> edges).<br />
<br />
'''Permutation:''' Let <math>z</math> be a set of node and edge features. Given a permutation <math>\sigma</math> of <math>\{1,\dots,n\}</math>, let <math>\sigma(z)</math> be a new set of node and edge features given by [<math>\sigma(z)]_i = z_{\sigma(i)}</math> and <math>[\sigma(z)]_{i,j} = z_{\sigma(i), \sigma(j)}</math><br />
<br />
'''One-hot representation:''' <math>\mathbf{1}[j]</math> be a one-hot vector with 1 in the <math>j^{th}</math> coordinate<br />
<br />
=Permutation-Invariant Structured prediction=<br />
<br />
With permutation-invariant structured prediction, we would expect the algorithm to produce the same result given the same score function. For instance, consider the case where we have label space for 3 variables <math>y_1, y_2, y_3</math> with input <math>\mathbf{z} = (f_1, f_2, f_3, f_{12}, f_{13}, f_{23})</math> that outputs label <math>\mathbf{y} = (y_1^*, y_2^*, y_3^*)</math>. Then if the algorithm is run on a permuted version input <math>z' = (f_2, f_1, f_3, f_{21}, f_{23}, f_{13})</math>, we would expect <math>\mathbf{y} = (y_2^*, y_1^*, y_3^*)</math> given the same score function.<br />
<br />
'''Graph permutation invariance (GPI):''' a graph labeling function <math>\mathcal{F}</math> is graph-permutation invariant, if for all permutations <math>\sigma</math> of <math>\{1, \dots, n\}</math> and for all nodes <math>z</math>, <math>\mathcal{F}(\sigma(\mathbf{z})) = \sigma(\mathcal{F}(\mathbf{z}))</math>. Practically speaking, graph permutation means that the same graph is constructed, no matter the order in which elements are predicted. In scene graph generation approaches, Region Proposal Networks are often used as an initial pre-processing step. The results from these (cropped images representing bounding boxes) are then sequentially fed through a respective vertex (or edge) detection network. The idea behind Permutation Invariance is that, no matter the order these are passed in, the final scene graph is identical. In effect, this means not connecting vertices that should not be connected simply because a more promising vertex has not yet been identified. <br />
<br />
The paper presents a theorem on the necessary and sufficient conditions for a function <math>\mathcal{F}</math> to be graph permutation invariant. Intuitively, because <math>\mathcal{F}</math> is a function that takes an ordered set <math>z</math> as input, the output on <math>\mathbf{z}</math> could very well be different from <math>\sigma(\mathbf{z})</math>, which means <math>\mathcal{F}</math> needs to have some sort of symmetry in order to sustain <math>[\mathcal{F}(\sigma(\mathbf{z}))]]_k = [\mathcal{F}(\mathbf{z})]_{\sigma(k)}</math>.<br />
<br />
[[File:graph_permutation_invariance.jpg|400px|center]]<br />
<br />
==Theorem 1==<br />
Let <math>\mathcal{F}</math> be a graph labeling function. Then <math>\mathcal{F}</math> is graph-permutation invariant if and only if there exist functions <math>\alpha, \rho, \phi</math> such that for all <math>k=1, .., n</math>:<br />
\begin{align}<br />
[\mathcal{F}(\mathbf{z})]_k = \rho(\mathbf{z}_k, \sum_{i=1}^n \alpha(\mathbf{z}_i, \sum_{i\neq j} \phi(\mathbf{z}_i, \mathbf{z}_{i,j}, \mathbf{z}_j)))<br />
\end{align}<br />
where <math>\phi: \mathbb{R}^{2d+e} \rightarrow \mathbb{R}^L, \alpha: \mathbb{R}^{d + L} \rightarrow \mathbb{R}^{W}, p: \mathbb{R}^{W+d} \rightarrow \mathbb{R}</math>.<br />
<br />
Notice that for the dimensions of inputs and outputs, <math>d</math> refers to the number of singleton features in <math>z</math> and <math>e</math> refers to the number of edges. <br />
<br />
[[File:GPI_architecture.jpg|thumb|A schematic representation of the GPI architecture. Singleton features <math>z_i</math> are omitted for simplicity. First, the features <math>z_{i,j}</math> are processed element-wise by <math>\phi</math>. Next, they are summed to create a vector <math>s_i</math>, which is concatenated with <math>z_i</math>. Third, a representation of the entire graph is created by applying <math>\alpha\ n</math> times and summing the created vector. The graph representation is then finally processed by <math>\rho</math> together with <math>z_k</math>.|600px|center]]<br />
<br />
==Proof Sketch for Theorem 1==<br />
The proof of this theorem can be found in the paper. A proof sketch is provided below:<br />
<br />
'''For the forward direction''' (function that follows the form set out in equation (1) is GPI):<br />
# Using definition of permutation <math>\sigma</math>, and rewriting <math>[F(z)]_{\sigma(k)}</math> in the form from equation (1)<br />
# Second argument of <math>\rho</math> is invariant under <math>\sigma</math>, since it takes the sum of all indices <math>i</math> and all other indices <math>j \neq i </math>.<br />
<br />
'''For the backward direction''' (any black-box GPI function can be expressed in the form of equation 1):<br />
# Construct <math>\phi, \alpha</math> such that second argument of <math>\rho</math> contains all information about graph features of <math>z</math>, including edges that the features originate from<br />
# Assume each <math>z_k</math> uniquely identifies the node and <math>\mathcal{F}</math> is a function only of pairwise features <math>z_{i,j}</math><br />
# Construct <math>H</math> be a perfect hash function with <math>L</math> buckets, and <math>\phi</math> which maps '''pairwise features''' to a vector of size <math>L</math><br />
# <math>*</math>Construct <math>\phi(z_i, z_{i,j}, z_j) = \mathbf{1}[H(z_j)] z_{i,j}</math>, which intuitively means that <math>\phi</math> stores <math>z_{i,j}</math> in the unique bucket for node <math>j</math><br />
# Construct function <math>\alpha</math> to output a matrix <math>\mathbb{R}^{L \times L}</math> that maps each pairwise feature into unique positions (<math>\alpha(z_i, s_i) = \mathbf{1}[H(z_i)]s_i^T</math>)<br />
# Construct matrix <math>M = \sum_i \alpha(z_i,s_i)</math> by discarding rows/columns in <math>M</math> that do not correspond to original nodes (which reduces dimension to <math>n\times n</math>; set <math>\rho</math> to have same outcome as <math>\mathcal{F}</math>, and set the output of <math>\mathcal{F}</math> on <math>M</math> to be the labels <math>\mathbf{y} = y_1, \dots, y_n</math><br />
<br />
<math>*</math>The paper presents the proof for the edge features <math>z_{ij}</math> being scalar (<math>e = 1</math>) for simplicity, which can be extended easily to vectors with additional indexing.<br />
<br />
Although the results discussed previously apply to complete graphs (edges apply to all feature pairs), it can be easily extended to incomplete graphs. For incomplete graphs, the input to F only contains the features corresponding to valid edges of the graph. The authors are only interested in invariances that preserve the graph structure. Thus, in place of permutation-invariance, it is now an automorphism-invariance.<br />
<br />
==Implications and Applications of Theorem 1==<br />
===Key Implications of Theorem 1===<br />
# Architecture "collects" information from the different edges of the graph, and does so in an invariant fashion using <math>\alpha</math> and <math>\phi</math><br />
# Architecture is parallelizable, since all <math>\phi</math> functions can be applied simultaneously<br />
<br />
===Some applications of Theorem 1===<br />
# '''Attention:''' the concept of attention can be implemented in the GPI characterization, with slight alterations to the functions <math>\alpha</math> and <math>\phi</math>. In attention each node aggregates features of neighbours through a function of neighbour's relevance. Which means the lable of an entity could depend strongly on its close entity. The complete details can be found in the supplementary materials of the paper.<br />
<br />
# '''RNN:''' recurrent architectures can maintain GPI property, since all GPI function <math>\mathcal{F}</math> are closed under composition. The output of one step after running <math>\mathcal{F}</math> will act as input for the next step, but maintain the GPI property throughout.<br />
<br />
=Related Work=<br />
# '''Architectural invariance:''' suggested recently in a 2017 paper called Deep Sets by Zaheer et al., which considers the case of invariance that is more restrictive.<br />
# '''Deep structured prediction:''' previous work applied deep learning to structured prediction, for instance, semantic segmentation. Some algorithms include message passing algorithms, gradient descent for maximizing score functions, greedy decoding (inference of labels based on time of previous labels). For example, Xu et al. 2017 proposes a novel end-to-end model that generates structured scene representation, and their model solves the scene graph inference problem using standard RNNs and learns to iteratively improves its predictions via message passing. Apart from those algorithms, deep learning has been applied to other graph-based problems such as the Travelling Salesman Problem (Bello et al., 2016; Gilmer et al., 2017; Khalil et al., 2017). However, none of the previous work specifically address the notion of invariance in the general architecture, but rather focus on message passing architectures that can be generalized by this paper.<br />
# '''Scene graph prediction:''' scene graph extraction allows for reasoning, question answering, and image retrieval (Johnson et al., 2015; Lu et al., 2016; Raposo et al., 2017). Some other works in this area include object detection, action recognition, and even detection of human-object interactions (Liao et al., 2016; Plummer et al., 2017). Additional work has been done with the use of message passing algorithms (Xu et al., 2017), word embeddings (Lu et al., 2016), and end-to-end prediction directly from pixels (Newell & Deng, 2017). A notable mention is NeuralMotif (Zellers et al., 2017), which the authors describe as the current state-of-the-art model for scene graph predictions on Visual Genome dataset.<br />
# '''Burst Image Deblurring Using Permutation Invariant Convolutional Neural Networks:''' similar ideas were applied, where Permutation Invariant CNN, are used to restore sharp and noise-free images from bursts of photographs affected by hand tremor and noise. This presented good quality images with lots of details for challenging datasets.<br />
<br />
=Experimental Results=<br />
<br />
The authors evaluated the advantage of GPI architectures empirically. They first utilized synthetic graph labeling and then used scene-graph classification for mapping images.<br />
<br />
==Synthetic Graph Labeling==<br />
The authors created a synthetic problem to study GPI. This involved using an input graph <math>G = (V,E)</math> where each node <math>i</math> belongs to the set <math>\Gamma(i) \in \{1, \dots, K\}</math> where <math>K</math> is the number of samples. The task is to compute for each node, the number of neighbours that belong to the same set (i.e. finding the label of the node <math>i</math> if <math>y_i = \sum_{j \in N(i)} \mathbf{1}[\Gamma(i) = \Gamma(j)]</math>) . Then, random graphs (each with 10 nodes) were generated by sampling edges, and the set <math>\Gamma(i) \in \{1, \dots, K\}</math>for each node independently and uniformly.<br />
The node features of the graph <math>z_i \in \{0,1\}^K</math> are one-hot vectors of <math>\Gamma(i)</math>, and each pairwise edge feature <math>z_{ij} \in \{0, 1\}</math> denote whether the edge <math>ij</math> is in the edge set <math>E</math>. <br />
3 architectures were studied in this paper:<br />
# '''GPI-architecture for graph prediction''' (without attention and RNN)<br />
# '''LSTM''': replacing <math>\sum \phi(\cdot)</math> and <math>\sum \alpha(\cdot)</math> in the form of Theorem 1 using two LSTMs with state size 200, reading their input in random order<br />
# '''Fully connected feed-forward network''': with 2 hidden layers, each layer containing 1,000 nodes; the input is a concatenation of all nodes and pairwise features, and the output is all node predictions<br />
<br />
The results show that the GPI architecture requires far fewer samples to converge to the correct solution.<br />
[[File:GPI_synthetic_example.jpg|450px|center]]<br />
<br />
This experimental result is meant to demonstrate sample complexity. For fairness, all three models were constructed with a similar number of trainable parameters. The results tie back in with the author's comment that a black-box model which violates permutation invariant structure wastes capacity on learning it at training time. This illustrates the advantage of an architecture with a proper inductive bias.<br />
<br />
==Scene-Graph Classification==<br />
Applying the concept of GPI to Scene-Graph Prediction (SGP) is the main task of this paper. The input to this problem is an image, along with a set of annotated bounding boxes for the entities in the image. The goal is to correctly label each entity within the bounding boxes and the relationship between every pair of entities, resulting in a coherent scene graph.<br />
<br />
The authors describe two different types of variables to predict. The first type is entity variables <math>[y_1, \dots, y_n]</math> for all bounding boxes, where each <math>y_i</math> can take one of L values and refers to objects such as "dog" or "man". The second type is relation variables <math>[y_{n+1}, \cdots, y_{n^2}]</math>, where each <math>y_i</math> represents the relation (e.g. "on", "below") between a pair of bounding boxes (entities).<br />
<br />
The scene graph and contain two types of edges:<br />
# '''Entity-entity edge''': connecting two entities <math>y_i</math> and <math>y_j</math> for <math>1 \leq i \neq j \leq n</math><br />
# '''Entity-relation edges''': connecting every relation variable <math>y_k</math> for <math>k > n</math> to two entities<br />
<br />
The feature set <math>\mathbf{z}</math> is based on the baseline model from Zellers et al. (2017). For entity variables <math>y_i</math>, the vector <math>\mathbf{z}_i \in \mathbb{R}^L</math> models the probability of the entity appearing in <math>y_i</math>. <math>\mathbf{z}_i</math> is augmented by the coordinates of the bounding box. Similarly for relation variables <math>y_j</math>, the vector <math>\mathbf{z}_j \in \mathbb{R}^R</math>, models the probability of the relations between the two entities in <math>j</math>. For entity-entity pairwise features <math>\mathbf{z}_{i,j}</math>, there is a similar representation of the probabilities for the pair. The SGP outputs probability distributions over all entities and relations, which will then be used as input recurrently to maintain GPI. Finally, word embeddings are used and concatenated for the most probable entity-relation labels.<br />
<br />
'''Components of the GPI architecture''' (ent for entity, rel for relation)<br />
# <math>\phi_{ent}</math>: network that integrates two entity variables <math>y_i</math> and <math>y_j</math>, with input <math>z_i, z_j, z_{i,j}</math> and output vector of <math>\mathbb{R}^{n_1}</math> <br />
# <math>\alpha_{ent}</math>: network with inputs from <math>\phi_{ent}</math> for all neighbours of an entity, and uses attention mechanism to output vector <math>\mathbb{R}^{n_2}</math> <br />
# <math>\rho_{ent}</math>: network with inputs from the various <math>\mathbb{R}^{n_2}</math> vectors, and outputs <math>L</math> logits to predict entity value<br />
# <math>\rho_{rel}</math>: network with inputs <math>\alpha_{ent}</math> of two entities and <math>z_{i,j}</math>, and output into <math>R</math> logits<br />
<br />
==Set-up and Results==<br />
'''Dataset''': based on Visual Genome (VG) by (Krishna et al., 2017), which contains a total of 108,077 images annotated with bounding boxes, entities, and relations. An average of 12 entities and 7 relations exist per image. For a fair comparison with previous works, data from (Xu et al., 2017) for train and test splits were used. The authors used the same 150 entities and 50 relations as in (Xu et al., 2017; Newell & Deng, 2017; Zellers et al., 2017). Hyperparameters were tuned using a 70K/5K/32K split for training, validation, and testing respectively.<br />
<br />
'''Training''': all networks were trained using the Adam optimizer, with a batch size of 20. The loss function was the sum of cross-entropy losses over all of entities and relations. Penalties for misclassified entities were 4 times stronger than that of relations. Penalties for misclassified negative relations were 10 times weaker than that of positive relations.<br />
<br />
'''Evaluation''': there are three major tasks when inferring from the scene graph. The authors focus on the following:<br />
# '''SGCIs''': given ground-truth entity bounding boxes, predict all entity and relations categories<br />
# '''PredCIs''': given annotated bounding boxes with entity labels, predict all relations<br />
<br />
The evaluation metric Recall@K (shortened to R@K) is drawn from (Lu et al., 2016). This metric is the fraction of correct ground-truth triplets that appear within the <math>K</math> most confident triplets predicted by the model. Graph-constrained protocol requires the top-<math>K</math> triplets to assign one consistent class per entity and relation. The unconstrained protocol does not enforce such constraint.<br />
<br />
'''Models and baselines''': The authors compared variants of the GPI approach against four baselines, state-of-the-art models on completing scene graph sub-tasks. To maintain consistency, all models used the same training/testing data split, in addition to the preprocessing as per (Xu et al., 2017).<br />
<br />
'''Baselines from existing state-of-the-art models'''<br />
# (Lu et al., 2016): use of word embeddings to fine-tune the likelihood of predicted relations<br />
# (Xu et al., 2017): message passing algorithm between entities and relations to iteratively improve feature map for prediction<br />
# (Newell & Deng, 2017): Pixel2Graph, uses associative embeddings to produce a full graph from image<br />
# (Zellers et al., 2017): NeuralMotif method, encodes global context to capture higher-order motif in scene graphs; Baseline outputs entities and relations distributions without using global context<br />
<br />
'''GPI models'''<br />
# '''GPI with no attention mechanism''': simply following Theorem 1's functional form, with summation over features<br />
# '''GPI NeighborAttention''': same GPI model, but considers attention over neighbours features<br />
# '''GPI Linguistic''': similar to NeighborAttention model, but concatenates word embedding vectors<br />
<br />
'''Key Results''': The GPI Linguistic approach outperforms all baseline for SGCIs, and has similar performance to the state of the art NeuralMotifs method. The authors argue that PredCI is an easier task with less structure, yielding high performance for the existing state of the art models.<br />
<br />
[[File:GPI_table_results.png|700px|center]]<br />
<br />
=Conclusion=<br />
<br />
A deep learning approach was presented in this paper to structured prediction, which constrains the architecture to be invariant to structurally identical inputs. This approach relies on pairwise features which are capable of describing inter-label correlations and inherits the intuitive aspect of score-based approaches. The output produced is invariant to equivalent representation of the pairwise terms. <br />
<br />
As future work, the axiomatic approach can be extended; for example in image labeling, geometric variances such as shifts or rotations may be desired (or in other cases invariance to feature permutations may be desired). Additionally, exploring algorithms that discover symmetries for deep structured prediction when invariant structure is unknown and should be discovered from data is also an interesting extension of this work.<br />
<br />
=Critique=<br />
The paper's contribution comes from the novelty of the permutation invariance as a design guideline for structured prediction. Although not explicitly considered in many of the previous works, the idea of invariance in architecture has already been considered in Deep Sets by (Zaheer et al., 2017). This paper characterizes relaxes the condition on the invariance as compared to that of previous works. In the evaluation of the benefit of GPI models, the paper used a synthetic problem to illustrate the fact that far fewer samples are required for the GPI model to converge to 100% accuracy. However, when comparing the true task of scene graph prediction against the state-of-the-art baselines, the GPI variants had only marginal higher Recall@K scores. The true benefit of this paper's discovery is the avoidance of maximizing a score function (leading computationally difficult problem), and instead directly producing output invariant to how we represent the pairwise terms.<br />
<br />
=References=<br />
<br />
[Lu et al., 2016] Lu, Cewu, Krishna, Ranjay, Bernstein, Michael S., and Li, Fei-Fei. Visual relationship detection with<br />
language priors. In European Conf. Comput. Vision, pp. 852–869, 2016.<br />
<br />
Roei Herzig, Moshiko Raboh, Gal Chechik, Jonathan Berant, Amir Globerson, Mapping Images to Scene Graphs with Permutation-Invariant Structured Prediction, 2018.<br />
<br />
Additional resources from Moshiko Raboh's [https://github.com/shikorab/SceneGraph GitHub]</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=A_Bayesian_Perspective_on_Generalization_and_Stochastic_Gradient_Descent&diff=41057A Bayesian Perspective on Generalization and Stochastic Gradient Descent2018-11-23T01:43:32Z<p>Gchalato: /* Contribution */</p>
<hr />
<div>==Introduction==<br />
This work builds on Zhang et al.(2016), who showed deep neural networks can easily memorize randomly labeled training data, despite generalizing well on real labels of the same inputs. The authors consider two questions: how can we predict if a minimum will generalize to the test set, and why does stochastic gradient descent find minima that generalize well? <br />
They show that the same phenomenon occurs in small linear models. These observations are explained by the Bayesian evidence, which penalizes sharp minima but is invariant to model parameterization. We also demonstrate that, when one holds the learning rate fixed, there is an optimum batch size which maximizes the test set accuracy.<br />
We propose that the noise introduced by small mini-batches drives the parameters towards minima whose evidence is large. Interpreting stochastic gradient descent as a stochastic differential equation, we identify the “noise scale” <math display="inline"> g \approx \epsilon N/B </math> where <math display="inline">ε</math> is the learning rate, <math display="inline">N</math> the training set size and <math display="inline">B</math> the batch size. Consequently the optimum batch size is proportional to both the learning rate and the size of the training set, <math display="inline">B_{opt} \propto \epsilon N</math> . They verify these predictions empirically.<br />
<br />
==Motivation==<br />
This paper shows Bayesian principles can explain many recent observations in the deep learning literature, while also discovering practical new insights. Zhang et al. (2016) trained deep convolutional networks on ImageNet and CIFAR10, achieving excellent accuracy on both training and test sets. They then took the same input images, but randomized the labels, and found that while their networks were now unable to generalize to the test set, they still memorized the training labels. They claimed these results contradict learning theory, although this claim is disputed (Kawaguchi et al., 2017; Dziugaite & Roy, 2017). Nonetheless, their results beg the question; if our models can assign arbitrary labels to the training set, why do they work so well in practice? Meanwhile, Keskar et al. (2016) observed that if we hold the learning rate fixed and increase the batch size, the test accuracy usually falls. This striking result shows improving our estimate of the full-batch gradient can harm performance. Goyal et al. (2017) observed a linear scaling rule between batch size and learning rate in a deep ResNet, while Hoffer et al. (2017) proposed a square root rule on theoretical grounds.<br />
Many authors have suggested “broad minima” whose curvature is small may generalize better than “sharp minima” whose curvature is large (Chaudhari et al., 2016; Hochreiter & Schmidhuber, 1997). Indeed, Dziugaite & Roy (2017) argued the results of Zhang et al. (2016) can be understood using “nonvacuous” PAC-Bayes generalization bounds which penalize sharp minima, while Keskar et al. (2016) showed stochastic gradient descent (SGD) finds wider minima as the batch size is reduced. However, Dinh et al. (2017) challenged this interpretation, by arguing that the curvature of a minimum can be arbitrarily increased by changing the model parameterization.<br />
<br />
==Contribution==<br />
<br />
The main contributions of this paper are to show that:<br />
* The results of Zhang et al. (2016) are not unique to deep learning; it is observed the same phenomenon in a small “over-parameterized” linear model. Overparameterization occurs when a model is able to effectively “remember” training data. This occurs when there are enough parameters that the system of equations ends up with an infinite number of possible solutions. One can see why this over-training would lead to poor results in test cases, as this “memorization” learns noise as opposed to the inherent structure of different classes. It is demonstrated that this phenomenon is straightforwardly understood by evaluating the Bayesian evidence in favor of each model, which penalizes sharp minima but is invariant to the model parameterization.<br />
* SGD integrates a stochastic differential equation whose “noise scale” <math>g &asymp; &epsilon;N/B</math>, where<br />
&epsilon; is the learning rate, <math>N</math> training set size and <math>B</math> batch size. Noise drives SGD away from sharp minima, and therefore there is an optimal batch size which maximizes the test set accuracy. This optimal batch size is '''proportional to the learning rate and training set size'''.<br />
<br />
Zhang et al. (2016) showed high training competency of neural networks under informative labels, but drastic overfitting on improper labels. This implies weak generalizability even when a small proportion of labels are improper. The authors show that generalization is strongly correlated with the Bayesian evidence, a weighted combination of the depth of a minimum (the cost function) and its breadth (the Occam factor). Bayesians tend to make distributional assumptions on gradient updates by adding isotropic Gaussian noise. This paper builds upon these Bayesian principles by driving SGD away from sharp minima, and towards broad minima (the more broad, the better generalization due to less influence from small perturbations within input). The stochastic differential equation used as a component of gradient updates effectively serves as injected noise that improves a network's generalizability.<br />
<br />
==Main Results==<br />
<br />
The weakly regularized model memorizes random labels, however, generalizes properly on informative labels. Besides, the predictions are overconfident. The authors also showed that the test accuracy peaks at an optimal batch size, if one holds the other SGD hyper-parameters constant. It is postulated that the optimum represents a tradeoff between depth and breadth in the Bayesian evidence. However it is the underlying scale of random fluctuations in the SGD dynamics which controls the tradeoff, not the batch size itself. Furthermore, this test accuracy peak shifts as the training set size rises. The authors observed that the best found batch size is proportional to the learning rate. This scaling rule allowed the authors to increase the learning rate by simultaneously increasing the batch size with no loss in test accuracy and no increase in computational cost, thus parallelism across multiple GPU's can be fully leveraged to easily decrease training time. The scaling rule could also be applied to production models by consequentially increasing the batch size as new training data is introduced.<br />
<br />
==Conclusion==<br />
<br />
The paper showed that Mini-batch noise helps SGD to go away from sharp minima, and provided an evidence that there is an optimal optimum batch size for a maximum the test accuracy. Based on interpreting SGD as integrating stochastic differential equation, this batch size is proportional to the learning rate and the training set size. Moreover, the authors shown that Bopt ∝ 1/(1 − m), where m is the momentum coefficient. More analysis was done on the relation between the learning rate, effective learning rate, and batch size is presented in ICLR 2018, where the authors proved by experiments that all the benefits of decaying the learning rate are achieved by increasing the batch size in addition to reducing the number of parameter updates dramatically, and also were able use literature parameters without the need of any hyper parameter tuning (Samuel L. Smith, Pieter-Jan Kindermans, Chris Ying, Quoc V. Le).<br />
<br />
==References==<br />
Chaudhari, Pratik, et al. "Entropy-sgd: Biasing gradient descent into wide valleys." arXiv preprint arXiv:1611.01838 (2016).<br />
<br />
Dziugaite, Gintare Karolina, and Daniel M. Roy. "Computing nonvacuous generalization bounds for deep (stochastic) neural networks with many more parameters than training data." arXiv preprint arXiv:1703.11008 (2017).<br />
<br />
Germain, Pascal, et al. "Pac-bayesian theory meets bayesian inference." Advances in Neural Information Processing Systems. 2016.<br />
<br />
Goyal, Priya, et al. "Accurate, large minibatch SGD: training imagenet in 1 hour." arXiv preprint arXiv:1706.02677 (2017).<br />
<br />
Gull, Stephen F. "Bayesian inductive inference and maximum entropy." Maximum-entropy and Bayesian methods in science and engineering. Springer, Dordrecht, 1988. 53-74.<br />
<br />
Hoffer, Elad, Itay Hubara, and Daniel Soudry. "Train longer, generalize better: closing the generalization gap in large batch training of neural networks." Advances in Neural Information Processing Systems. 2017.<br />
Kass, Robert E., and Adrian E. Raftery. "Bayes factors." Journal of the american statistical association 90.430 (1995): 773-795.<br />
<br />
Kawaguchi, Kenji, Leslie Pack Kaelbling, and Yoshua Bengio. "Generalization in deep learning." arXiv preprint arXiv:1710.05468 (2017).<br />
<br />
Keskar, Nitish Shirish, et al. "On large-batch training for deep learning: Generalization gap and sharp minima." arXiv preprint arXiv:1609.04836 (2016).<br />
<br />
MacKay, David JC. "A practical Bayesian framework for backpropagation networks." Neural computation 4.3 (1992): 448-472.<br />
<br />
Zhang, Chiyuan, et al. "Understanding deep learning requires rethinking generalization." arXiv preprint arXiv:1611.03530 (2016).</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=stat946w18/Wavelet_Pooling_For_Convolutional_Neural_Networks&diff=41054stat946w18/Wavelet Pooling For Convolutional Neural Networks2018-11-23T01:23:06Z<p>Gchalato: /* Background */</p>
<hr />
<div>=Wavelet Pooling For Convolutional Neural Networks=<br />
<br />
[https://goo.gl/forms/8NucSpF36K6IUZ0V2 Your feedback on presentations]<br />
<br />
<br />
== Introduction, Important Terms and Brief Summary==<br />
<br />
This paper focuses on the following important techniques: <br />
<br />
1) Convolutional Neural Nets (CNN): These are networks with layered structures that conform to the shape of inputs rather than vector-based features and consistently obtain high accuracies in the classification of images and objects. Researchers continue to focus on CNN to improve their performances. <br />
<br />
2) Pooling: Pooling subsamples the results of the convolution layers and gradually reduces spatial dimensions of the data throughout the network. It is done to reduce parameters, increase computational efficiency and regulate overfitting. <br />
<br />
Some of the pooling methods, including max pooling and average pooling, are deterministic. Deterministic pooling methods are efficient and simple, but can hinder the potential for optimal network learning. In contrast, mixed pooling and stochastic pooling use a probabilistic approach, which can address some problems of deterministic methods. The neighborhood approach is used in all the mentioned pooling methods due to its simplicity and efficiency. Nevertheless, the approach can cause edge halos, blurring, and aliasing which need to be minimized. This paper introduces wavelet pooling, which uses a second-level wavelet decomposition to subsample features. The nearest neighbor interpolation is replaced by an organic, subband method that more accurately represents the feature contents with fewer artifacts. The method decomposes features into a second level decomposition and discards first level subbands to reduce feature dimensions. This method is compared to other state-of-the-art pooling methods to demonstrate superior results. Tests are conducted on benchmark classification tests like MNIST, CIFAR10, SHVN and KDEF.<br />
<br />
For further information on wavelets, follow this link to MathWorks' [https://www.mathworks.com/videos/understanding-wavelets-part-1-what-are-wavelets-121279.html Understanding Wavelets] video series.<br />
<br />
== Intuition ==<br />
<br />
Convolutional networks commonly employ convolutional layers to extract features and use pooling methods for spatial dimensionality reduction. In this study, wavelet pooling is introduced as an alternative to traditional neighborhood pooling by providing a more structural feature dimension reduction method. Max pooling is addressed to have over-fitting problems and average pooling is mentioned to smooth out or 'dilute' details in features.<br />
<br />
Pooling is often introduced within networks to ensure local invariance to prevent overfitting due to small transitional shifts within an image. Despite the effectiveness of traditional pooling methods such as max pooling introduce this translational invariance by discarding information using methods analogous to nearest neighbour interpolation. With the hope of providing a more organic way of pooling, the authors leverage all information within cells inputted within a pooling operation with the hope that the resulting dim-reduced features are able to contain information from all high level cells using various dot products.<br />
<br />
== History ==<br />
<br />
A history of different pooling methods have been introduced and referenced in this study:<br />
* manual subsampling at 1979<br />
* Max pooling at 1992<br />
* Mixed pooling at 2014<br />
* pooling methods with probabilistic approaches at 2014 and 2015<br />
<br />
== Background ==<br />
Average Pooling and Max Pooling are well-known pooling methods and are popular techniques used in the literature. These pooling methods reduce input data dimensionality by taking the maximum value or the average value of specific areas and condense them into one single value. While these methods are simple and effective, they still have some limitations. The authors identify the following limitations:<br />
<br />
'''Limitations of Max Pooling and Average Pooling'''<br />
<br />
'''Max pooling''': takes the maximum value of a region <math>R_{ij} </math> and selects it to obtain a condensed feature map. It can '''erase the details''' of the image (happens if the main details have less intensity than the insignificant details) and also commonly '''over-fits''' the training data. The max-pooling is defined as:<br />
<br />
\begin{align}<br />
a_{kij} = max_{(p,q)\in R_{ij}}(a_{kpq})<br />
\end{align}<br />
<br />
'''Average pooling''': calculates the average value of a region and selects it to obtain a condensed feature map. Depending on the data, this method can '''dilute pertinent details''' from an image (happens for data with values much lower than the significant details) The avg-pooling is defined as:<br />
<br />
\begin{align}<br />
a_{kij} = \frac{1}{|R_{ij}|}\sum_{(p,q)\in R_{ij}}{{a_{kpq}}}<br />
\end{align}<br />
<br />
Where <math>a_{kij}</math> is the output activation of the <math>k^{th}</math> feature map at <math>(i,j)</math>, <math>a_{kpq}</math> is the input activation at<br />
<math>(p,q)</math> within <math>R_{ij}</math>, and <math>|R_{ij}|</math> is the size of the pooling region. Figure 2 provides an example of the weaknesses of these two methods using toy images:<br />
<br />
[[File: fig0001.PNG| 700px|center]]<br />
<br />
<br />
'''How the researchers try to '''combat these issues'''?'''<br />
Using '''probabilistic pooling methods''' such as:<br />
<br />
1. '''Mixed pooling''': In general, when facing a new problem in which one would want to use a CNN, it is unintuitive to whether average or max-pooling is preferred. Notably, both techniques have significant drawbacks. Average pooling forces the network to consider low magnitude (and possibly irrelevant information) in constructing representations, while max pooling can force the network to ignore fundamental differences between neighbouring groups of pixels. To counteract this, mixed pooling probabilistically decides which to use during training / testing. It should be noted that, for training, it is only probabilistic in the forward pass. During back-propagation the network defaults to the earlier chosen method. Mixed pooling can be applied in 3 different ways.<br />
<br />
* For all features within a layer<br />
* Mixed between features within a layer<br />
* Mixed between regions for different features within a layer<br />
<br />
Mixed Pooling is defined as:<br />
<br />
\begin{align}<br />
a_{kij} = \lambda \cdot max_{(p,q)\in R_{ij}}(a_{kpq})+(1-\lambda) \cdot \frac{1}{|R_{ij}|}\sum_{(p,q)\in R_{ij}}{{a_{kpq}}}<br />
\end{align}<br />
<br />
Where <math>\lambda</math> is a random value 0 or 1, indicating max or average pooling.<br />
<br />
2. '''Stochastic pooling''': improves upon max pooling by randomly sampling from neighborhood regions based on the probability values of each activation. This is defined as:<br />
<br />
\begin{align}<br />
a_{kij} = a_l ~ \text{where } ~ l\sim P(p_1,p_2,...,p_{|R_{ij}|})<br />
\end{align}<br />
<br />
with probability of activations within each region defined as follows:<br />
<br />
\begin{align}<br />
p_{pq} = \dfrac{a_{pq}}{\sum_{(p,q)} \in R_{ij} a_{pq}}<br />
\end{align}<br />
<br />
The figure below describes the process of Stochastic Pooling. The figure on the left shows the activations of a given region, and the corresponding probability is shown in the center. The activations with the highest probability is selected by the pooling method. However, any activation can be selected. In this case, the midrange activation of 13% is selected. <br />
<br />
[[File: stochastic pooling.jpeg| 700px|center]]<br />
<br />
As stochastic pooling is based on probability and is not deterministic, it avoids the shortcomings of max and average pooling and enjoys some of the advantages of max pooling.<br />
<br />
3. "Top-k activation pooling" is the method that picks the top-k activation in every pooling region. This makes sure that the maximum information can pass through subsampling gates. It is to be used with max pooling, but after max pooling, to further improve the representation capability, they pick top-k activation, sum them up, and constrain the summation by a constant. <br />
Details in this paper: https://www.hindawi.com/journals/wcmc/2018/8196906/<br />
<br />
'''Wavelets and Wavelet Transform'''<br />
A wavelet is a representation of some signal. For use in wavelet transforms, they are generally represented as combinations of basis signal functions.<br />
<br />
The wavelet transform involves taking the inner product of a signal (in this case, the image), with these basis functions. This produces a set of coefficients for the signal. These coefficients can then be quantized and coded in order to compress the image.<br />
<br />
One issue of note is that wavelets offer a tradeoff between resolution in frequency, or in time (or presumably, image location). For example, a sine wave will be useful to detect signals with its own frequency, but cannot detect where along the sine wave this alignment of signals is occuring. Thus, basis functions must be chosen with this tradeoff in mind.<br />
<br />
Source: Compressing still and moving images with wavelets<br />
<br />
== Proposed Method ==<br />
<br />
The proposed pooling method uses wavelets (i.e. small waves - generally used in signal processing) to reduce the dimensions of the feature maps. They use wavelet transform to minimize artifacts resulting from neighborhood reduction. They postulate that their approach, which discards the first-order sub-bands, more organically captures the data compression.<br />
<br />
* '''Forward Propagation'''<br />
<br />
The proposed wavelet pooling scheme pools features by performing a 2nd order decomposition in the wavelet domain according to the fast wavelet transform (FWT) which is a more efficient implementation of the two-dimensional discrete wavelet transform (DWT) as follows:<br />
<br />
\begin{align}<br />
W_{\varphi}[j+1,k] = h_{\varphi}[-n]*W_{\varphi}[j,n]|_{n=2k,k\leq0}<br />
\end{align}<br />
<br />
\begin{align}<br />
W_{\psi}[j+1,k] = h_{\psi}[-n]*W_{\psi}[j,n]|_{n=2k,k\leq0}<br />
\end{align}<br />
<br />
where <math>\varphi</math> is the approximation function, and <math>\psi</math> is the detail function, <math>W_{\varphi},W_{\psi}</math> are called approximation and detail coefficients. <math>h_{\varphi[-n]}</math> and <math>h_{\psi[-n]}</math> are the time reversed scaling and wavelet vectors, (n) represents the sample in the vector, while (j) denotes the resolution level<br />
<br />
When using the FWT on images, it is applied twice (once on the rows, then again on the columns). By doing this in combination, the detail sub-bands (LH, HL, HH) at each decomposition level, and approximation sub-band (LL) for the highest decomposition level is obtained.<br />
After performing the 2nd order decomposition, the image features are reconstructed, but only using the 2nd order wavelet sub-bands. This method pools the image features by a factor of 2 using the inverse FWT (IFWT) which is based off the inverse DWT (IDWT).<br />
<br />
\begin{align}<br />
W_{\varphi}[j,k] = h_{\varphi}[-n]*W_{\varphi}[j+1,n]+h_{\psi}[-n]*W_{\psi}[j+1,n]|_{n=\frac{k}{2},k\leq0}<br />
\end{align}<br />
<br />
[[File: wavelet pooling forward.PNG| 700px|center]]<br />
<br />
<br />
* '''Backpropagation'''<br />
<br />
The proposed wavelet pooling algorithm performs backpropagation by reversing the process of its forward propagation. First, the image feature being backpropagated undergoes 1st order wavelet decomposition. After decomposition, the detail coefficient sub-bands up-sample by a factor of 2 to create a new 1st level decomposition. The initial decomposition then becomes the 2nd level decomposition. Finally, this new 2nd order wavelet decomposition reconstructs the image feature for further backpropagation using the IDWT. Figure 5, illustrates the wavelet pooling backpropagation algorithm in details:<br />
<br />
[[File:wavelet pooling backpropagation.PNG| 700px|center]]<br />
<br />
== Results and Discussion ==<br />
<br />
All experiments have been performed using the MatConvNet(Vedaldi & Lenc, 2015) architecture. Stochastic gradient descent has been used for training. For the proposed method, the Haar wavelet has been chosen as the basis wavelet for its property of having even, square sub-bands. All CNN structures except for MNIST use a network loosely based on Zeilers network (Zeiler & Fergus, 2013). The experiments are repeated with Dropout (Srivastava, 2013) and the Local Response Normalization (Krizhevsky, 2009) is replaced with Batch Normalization (Ioffe & Szegedy, 2015) for CIFAR-10 and SHVN (Dropout only) to examine how these regularization techniques change the pooling results. The authors have tested the proposed method on four different datasets as shown in the figure:<br />
<br />
[[File: selection of image datasets.PNG| 700px|center]]<br />
<br />
Different methods based on Max, Average, Mixed, Stochastic and Wavelet have been used at the pooling section of each architecture. Accuracy and Model Energy have been used as the metrics to evaluate the performance of the proposed methods. These have been evaluated and their performances have been compared on different data-sets.<br />
<br />
* MNIST:<br />
<br />
The network architecture is based on the example MNIST structure from MatConvNet, with batch-normalization, inserted. All other parameters are the same. The figure below shows their network structure for the MNIST experiments.<br />
<br />
[[File: CNN MNIST.PNG| 700px|center]]<br />
<br />
The input training data and test data come from the MNIST database of handwritten digits. The full training set of 60,000 images is used, as well as the full testing set of 10,000 images. The table below shows their proposed method outperforms all methods. Given the small number of epochs, max pooling is the only method to start to over-fit the data during training. Mixed and stochastic pooling show a rocky trajectory but do not over-fit. Average and wavelet pooling show a smoother descent in learning and error reduction. The figure below shows the energy of each method per epoch.<br />
<br />
[[File: MNIST pooling method energy.PNG| 700px|center]]<br />
<br />
<br />
The accuracies for both paradigms are shown below:<br />
<br />
<br />
[[File: MNIST perf.PNG| 700px|center]]<br />
<br />
* CIFAR-10:<br />
<br />
The authors perform two sets of experiments with the pooling methods. The first is a regular network structure with no dropout layers. They use this network to observe each pooling method without extra regularization. The second uses dropout and batch normalization and performs over 30 more epochs to observe the effects of these changes. <br />
<br />
[[File: CNN CIFAR.PNG| 700px|center]]<br />
<br />
The input training and test data come from the CIFAR-10 dataset. <br />
The full training set of 50,000 images is used, as well as the full testing set of 10,000 images. For both cases, with no dropout, and with dropout, Tables below show that the proposed method has the second highest accuracy.<br />
<br />
[[File: fig0000.jpg| 700px|center]]<br />
<br />
Max pooling over-fits fairly quickly, while wavelet pooling resists over-fitting. The change in learning rate prevents their method from over-fitting, and it continues to show a slower propensity for learning. Mixed and stochastic pooling maintain a consistent progression of learning, and their validation sets trend at a similar, but better rate than their proposed method. Average pooling shows the smoothest descent in learning and error reduction, especially in the validation set. The energy of each method per epoch is also shown below:<br />
<br />
[[File: CIFAR_pooling_method_energy.PNG| 700px|center]]<br />
<br />
<br />
* SHVN:<br />
<br />
Two sets of experiments are performed with the pooling methods. The first is a regular network structure with no dropout layers. They use this network to observe each pooling method without extra regularization same as what happened in the previous datasets.<br />
The second network uses dropout to observe the effects of this change. The figure below shows their network structure for the SHVN experiments:<br />
<br />
[[File: CNN SHVN.PNG| 700px|center]]<br />
<br />
The input training and test data come from the SHVN dataset. For the case with no dropout, they use 55,000 images from the training set. For the case with dropout, they use the full training set of 73,257 images, a validation set of 30,000 images they extract from the extra training set of 531,131 images, as well as the full testing set of 26,032 images. For both cases, with no dropout, and with dropout, Tables below show their proposed method has the second lowest accuracy.<br />
<br />
[[File: SHVN perf.PNG| 700px|center]]<br />
<br />
Max and wavelet pooling both slightly over-fit the data. Their method follows the path of max pooling but performs slightly better in maintaining some stability. Mixed, stochastic, and average pooling maintain a slow progression of learning, and their validation sets trend at near identical rates. The figure below shows the energy of each method per epoch.<br />
<br />
[[File: SHVN pooling method energy.PNG| 700px|center]]<br />
<br />
* KDEF:<br />
<br />
They run one set of experiments with the pooling methods that includes dropout. The figure below shows their network structure for the KDEF experiments:<br />
<br />
[[File:CNN KDEF.PNG| 700px|center]]<br />
<br />
The input training and test data come from the KDEF dataset. This dataset contains 4,900 images of 35 people displaying seven basic emotions (afraid, angry, disgusted, happy, neutral, sad, and surprised) using facial expressions. They display emotions at five poses (full left and right profiles, half left and right profiles, and straight).<br />
<br />
This dataset contains a few errors that they have fixed (missing or corrupted images, uncropped images, etc.). All of the missing images are at angles of -90, -45, 45, or 90 degrees. They fix the missing and corrupt images by mirroring their counterparts in MATLAB and adding them back to the dataset. They manually crop the images that need to match the dimensions set by the creators (762 x 562).<br />
KDEF does not designate a training or test data set. They shuffle the data and separate 3,900 images as training data, and 1,000 images as test data. They resize the images to 128x128 because of memory and time constraints.<br />
<br />
The dropout layers regulate the network and maintain stability in spite of some pooling methods known to over-fit. The table below shows their proposed method has the second highest accuracy. Max pooling eventually over-fits, while wavelet pooling resists over-fitting. Average and mixed pooling resist over-fitting but are unstable for most of the learning. Stochastic pooling maintains a consistent progression of learning. Wavelet pooling also follows a smoother, consistent progression of learning.<br />
The figure below shows the energy of each method per epoch.<br />
<br />
[[File: KDEF pooling method energy.PNG| 700px|center]]<br />
<br />
The accuracies for both paradigms are shown below:<br />
<br />
[[File: KDEF perf.PNG| 700px|center]]<br />
<br />
<br />
<br />
* Computational Complexity:<br />
Above experiments and implementations on wavelet pooling were more of a proof-of-concept rather than an optimized method. In terms of mathematical operations, the wavelet pooling method is the least computationally efficient compared to all other pooling methods mentioned above. Among all the methods, average pooling is the most efficient methods, max pooling and mix pooling are at a similar level while wavelet pooling is way more expensive to complete the calculation.<br />
<br />
== Conclusion ==<br />
<br />
They prove wavelet pooling has the potential to equal or eclipse some of the traditional methods currently utilized in CNNs. Their proposed method outperforms all others in the MNIST dataset, outperforms all but one in the CIFAR-10 and KDEF datasets, and performs within respectable ranges of the pooling methods that outdo it in the SHVN dataset. The addition of dropout and batch normalization show their proposed methods response to network regularization. Like the non-dropout cases, it outperforms all but one in both the CIFAR-10 & KDEF datasets and performs within respectable ranges of the pooling methods that outdo it in the SHVN dataset.<br />
<br />
== Suggested Future work ==<br />
<br />
Upsampling and downsampling factors in decomposition and reconstruction needs to be changed to achieve more feature reduction.<br />
The subbands that we previously discard should be kept for higher accuracies. To achieve higher computational efficiency, improving the FTW method is needed.<br />
<br />
== Critiques and Suggestions ==<br />
*The functionality of backpropagation process which can be a positive point of the study is not described enough comparing to the existing methods.<br />
* The main study is on wavelet decomposition while the reason of using Haar as mother wavelet and the number of decomposition levels selection has not been described and are just mentioned as a future study! <br />
* At the beginning, the study mentions that the pooling method is not under attention as it should be. In the end, results show that choosing the pooling method depends on the dataset and they mention trial and test as a reasonable approach to choose the pooling method. In my point of view, the authors have not really been focused on providing a pooling method which can help the current conditions to be improved effectively. At least, trying to extract a better pattern for relating results to the dataset structure could be so helpful.<br />
* Average pooling origins which are mentioned as the main pooling algorithm to compare with, is not even referenced in the introduction.<br />
* Combination of the wavelet, Max and Average pooling can be an interesting option to investigate more on this topic; both in a row(Max/Avg after wavelet pooling) and combined like mix pooling option.<br />
* While the current datasets express the performance of the proposed method in an appropriate way, it could be a good idea to evaluate the method using some larger datasets. Maybe it helps to understand whether the size of a dataset can affect the overfitting behavior of max pooling which is mentioned in the paper.<br />
* Adding asymptotic notations to the computational complexity of the proposed algorithm would be meaningful, particularly since the given results are for a single/fixed input size (one image in forward propagation) and consequently are not generalizable. <br />
* They could have considered comparing against Fast Fourier Transform (FFT). Including a non wavelet form seems to be an obvious candidate for comparison<br />
* If they went beyond the 2x2 pooling window this would have further supported their method<br />
* ([[https://openreview.net/forum?id=rkhlb8lCZ]]) The experiments are largely conducted with very small scale datasets. As a result, I am not sure if they are representative enough to show the performance difference between different pooling methods.<br />
* ([[https://openreview.net/forum?id=rkhlb8lCZ]]) No comparison to non-wavelet methods. For example, one obvious comparison would have been to look at using a DCT or FFT transform where the output would discard high-frequency components (this can get very close to the wavelet idea!).<br />
<br />
== References ==<br />
<br />
Williams, Travis, and Robert Li. "Wavelet Pooling for Convolutional Neural Networks." (2018).<br />
<br />
Hilton, Michael L., Björn D. Jawerth, and Ayan Sengupta. "Compressing still and moving images with wavelets." Multimedia systems 2.5 (1994): 218-227.<br />
<br />
<br />
== Revisions == <br />
<br />
*Two reviewers really liked the paper and one of them called it in the top 15% papers in the conference which supports the novelty and potential of the idea. One other reviewer, however, believed that this was not good enough to be accepted and the main reason for rejection was the linearity nature of wavelet(which was not convincingly described). <br />
<br />
*The main concern of two of the reviewers has been the size of the datasets that have been used to test the method and the authors have mentioned future works concerning bigger datasets to test the method.<br />
<br />
*The computational cost section has not been in the paper at the first place and was added after one of the reviewer's concern. So, the other reviewers have not been curious about this and unfortunately, there is no comment on that from them. However, the description on the non-efficient implementation seemed to be satisfactory to the reviewer which resulted in being accepted. <br />
<br />
[https://openreview.net/forum?id=rkhlb8lCZ Revisions]<br />
<br />
At the end, if you are interested in implementing the method, they are willing to share their code but after making it efficient. So, maybe there will be another paper regarding less computational cost on larger datasets with a publishable code.</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=CapsuleNets&diff=41030CapsuleNets2018-11-22T22:59:07Z<p>Gchalato: /* MultiMNIST */</p>
<hr />
<div>The paper "Dynamic Routing Between Capsules" was written by three researchers at Google Brain: Sara Sabour, Nicholas Frosst, and Geoffrey E. Hinton. This paper was published and presented at the 31st Conference on Neural Information Processing Systems (NIPS 2017) in Long Beach, California. The same three researchers recently published a highly related paper "Matrix Capsules with EM Routing" for ICLR 2018.<br />
<br />
=Motivation=<br />
<br />
Ever since AlexNet eclipsed the performance of competing architectures in the 2012 ImageNet challenge, convolutional neural networks have maintained their dominance in computer vision applications. Despite the recent successes and innovations brought about by convolutional neural networks, some assumptions made in these networks are perhaps unwarranted and deficient. Using a novel neural network architecture, the authors create CapsuleNets, a network that they claim is able to learn image representations in a more robust, human-like manner. With only a 3 layer capsule network, they achieved near state-of-the-art results on MNIST.<br />
==Adversarial Examples==<br />
<br />
First discussed by Christian Szegedy et. al. in late 2013, adversarial examples have been heavily discussed by the deep learning community as a potential security threat to AI learning. Adversarial examples are defined as inputs that an attacker creates intentionally fool a machine learning model. An example of an adversarial example is shown below: <br />
<br />
[[File:adversarial_img_1.png |center]]<br />
To the human eye, the image appears to be a panda both before and after noise is injected into the image, whereas the trained ConvNet model discerns the noisy image as a Gibbon with almost 100% certainty. The fact that the network is unable to classify the above image as a panda after the epsilon perturbation leads to many potential security risks in AI dependent systems such as self-driving vehicles. Although various methods have been suggested to combat adversarial examples, robust defences are hard to construct due to the inherent difficulties in constructing theoretical models for the adversarial example crafting process. However, beyond the fact that these examples may serve as a security threat, it emphasizes that these convolutional neural networks do not learn image classification/object detection patterns the same way that a human would. Rather than identifying the core features of a panda such as: its eyes, mouth, nose, and the gradient changes in its black/white fur, the convolutional neural network seems to be learning image representations in a completely different manner. Deep learning researchers often attempt to model neural networks after human learning, and it is clear that further steps must be taken to robustify ConvNets against targeted noise perturbations.<br />
<br />
==Drawbacks of CNNs==<br />
Hinton claims that the key fault with traditional CNNs lies within the pooling function. Although pooling builds translational invariance into the network, it fails to preserve spatial relationships between objects. When we pool, we effectively reduce a kxk kernel of convolved cells into a scalar input. This results in a desired local invariance without inhibiting the network's ability to detect features, but causes valuable spatial information to be lost.<br />
<br />
In the example below, the network is able to detect the similar features (eyes, mouth, nose, etc) within both images, but fails to recognize that one image is a human face, while the other is a Picasso-esque due to the CNN's inability to encode spatial relationships after multiple pooling layers.<br />
<br />
<br />
[[File:Equivariance Face.png |center]]<br />
<br />
Conversely, we hope that a CNN can recognize that both of the following pictures contain a kitten. Unfortunately, when we feed the two images into a ResNet50 architecture, only the first image is correctly classified, while the second image is predicted to be a guinea pig.<br />
<br />
<br />
[[File:kitten.jpeg |center]]<br />
<br />
<br />
[[File:kitten-rotated-180.jpg |center]]<br />
<br />
For a more in depth discussion on the problems with ConvNets, please listen to Geoffrey Hinton's talk "What is wrong with convolutional neural nets?" given at MIT during the Brain & Cognitive Sciences - Fall Colloquium Series (December 4, 2014).<br />
<br />
==Intuition for Capsules==<br />
Human vision ignores irrelevant details by using a carefully determined sequence of fixation points to ensure that only a tiny fraction of the optic array is ever processed at the highest resolution. Hinton argues that our brains reason visual information by deconstructing it into a hierarchical representation which we then match to familiar patterns and relationships from memory. The key difference between this understanding and the functionality of CNNs is that recognition of an object should not depend on the angle from which it is viewed. <br />
<br />
To enforce rotational and translational equivariance, Capsule Networks store and preserve hierarchical pose relationships between objects. The core idea behind capsule theory is the explicit numerical representations of relative relationships between different objects within an image. Building these relationships into the Capsule Networks model, the network is able to recognize newly seen objects as a rotated view of a previously seen object. For example, the below image shows the Statue of Liberty under five different angles. If a person had only seen the Statue of Liberty from one angle, they would be able to ascertain that all five pictures below contain the same object (just from a different angle).<br />
<br />
[[File:Rotational Invariance.jpeg |center]]<br />
<br />
Building on this idea of hierarchical representation of spatial relationships between key entities within an image, the authors introduce Capsule Networks. Unlike traditional CNNs, Capsule Networks are better equipped to classify correctly under rotational invariance. Furthermore, the authors managed to achieve state of the art results on MNIST using a fraction of the training samples that alternative state of the art networks require.<br />
<br />
<br />
=Background, Notation, and Definitions=<br />
<br />
==What is a Capsule==<br />
"Each capsule learns to recognize an implicitly defined visual entity over a limited domain of viewing conditions and deformations and it outputs both the probability that the entity is present within its limited domain and a set of “instantiation parameters” that may include the precise pose, lighting and deformation of the visual entity relative to an implicitly defined canonical version of that entity. When the capsule is working properly, the probability of the visual entity being present is locally invariant — it does not change as the entity moves over the manifold of possible appearances within the limited domain covered by the capsule. The instantiation parameters, however, are “equivariant” — as the viewing conditions change and the entity moves over the appearance manifold, the instantiation parameters change by a corresponding amount because they are representing the intrinsic coordinates of the entity on the appearance manifold."<br />
<br />
In essence, capsules store object properties in a vector form; probability of detection is encoded as the vector's length, while spatial properties are encoded as the individual vector components. Thus, when a feature is present but the image captures it under a different angle, the probability of detection remains unchanged.<br />
<br />
==Notation==<br />
<br />
We want the length of the output vector of a capsule to represent the probability that the entity represented by the capsule is present in the current input. The paper performs a non-linear squashing operation to ensure that vector length falls between 0 and 1, with shorter vectors (less likely to exist entities) being shrunk towards 0. <br />
<br />
\begin{align} \mathbf{v}_j &= \frac{||\mathbf{s}_j||^2}{1+ ||\mathbf{s}_j||^2} \frac{\mathbf{s}_j}{||\mathbf{s}_j||^2} \end{align}<br />
<br />
where <math>\mathbf{v}_j</math> is the vector output of capsule <math>j</math> and <math>s_j</math> is its total input.<br />
<br />
For all but the first layer of capsules, the total input to a capsule <math>s_j</math> is a weighted sum over all “prediction vectors” <math>\hat{\mathbf{u}}_{j|i}</math> from the capsules in the layer below and is produced by multiplying the output <math>\mathbf{u}i</math> of a capsule in the layer below by a weight matrix <math>\mathbf{W}ij</math><br />
<br />
\begin{align}<br />
\mathbf{s}_j = \sum_i c_{ij}\hat{\mathbf{u}}_{j|i}, \hat{\mathbf{u}}_{j|i}= \mathbf{W}_{ij}\mathbf{u}_i<br />
\end{align}<br />
where the <math>c_{ij}</math> are coupling coefficients that are determined by the iterative dynamic routing process.<br />
<br />
The coupling coefficients between capsule <math>i</math> and all the capsules in the layer above sum to 1 and are determined by a “routing softmax” whose initial logits <math>b_{ij}</math> are the log prior probabilities that capsule <math>i</math> should be coupled to capsule <math>j</math>.<br />
<br />
\begin{align}<br />
c_{ij} = \frac{\exp(b_{ij})}{\sum_k \exp(b_{ij})}<br />
\end{align}<br />
=Network Training and Dynamic Routing=<br />
<br />
==Understanding Capsules==<br />
The notation can get somewhat confusing, so I will provide intuition behind the computational steps within a capsule. The following image is taken from naturomic's talk on Capsule Networks.<br />
<br />
[[File:CapsuleNets.jpeg|center|800px]]<br />
<br />
The above image illustrates the key mathematical operations happening within a capsule (and compares them to the structure of a neuron). Although the operations are rather straightforward, it's crucial to note that the capsule computes an affine transformation onto each input vector. The length of the input vectors <math>\mathbf{u}_{i}</math> represent the probability of entity <math>i</math> existing in a lower level. This vector is then reoriented with an affine transform using <math>\mathbf{W}_{ij}</math> matrices that encode spatial relationships between entity <math>\mathbf{u}_{i}</math> and other lower level features.<br />
<br />
We illustrate the intuition behind vector-to-vector matrix multiplication within capsules using the following example: if vectors <math>\mathbf{u}_{1}</math>, <math>\mathbf{u}_{2}</math>, and <math>\mathbf{u}_{3}</math> represent detection of eyes, nose, and mouth respectively, then after multiplication with trained weight matrices <math>\mathbf{W}_{ij}</math> (where j denotes existence of a face), we should get a general idea of the general location of the higher level feature (face), similar to the image below.<br />
<br />
[[File:Predictions.jpeg |center]]<br />
<br />
==Dynamic Routing==<br />
A capsule <math>i</math> in a lower-level layer needs to decide how to send its output vector to higher-level capsules <math>j</math>. This decision is made with probability proportional to <math>c_{ij}</math>. If there are <math>K</math> capsules in the level that capsule <math>i</math> routes to, then we know the following properties about <math>c_{ij}</math>: <math>\sum_{j=1}^M c_{ij} = 1, c_{ij} \geq 0</math><br />
<br />
In essence, the <math>\{c_{ij}\}_{j=1}^M</math> denotes a discrete probability distribution with respect to capsule <math>i</math>'s output location. Lower level capsules decide which higher level capsules to send vectors into by adjusting the corresponding routing weights <math>\{c_{ij}\}_{j=1}^M</math>. After a few iterations in training, numerous vectors will have already been sent to all higher level capsules. Based on the similarity between the current vector being routed and all vectors already sent into the higher level capsules, we decide which capsule to send the current vector into.<br />
[[File:Dynamic Routing.png|center|900px]]<br />
<br />
In the image above, we notice that a cluster of points similar to the current vector has already been routed into capsule K, while most points in capsule J are high dissimilar. It thus makes more sense to route the current observation into capsule K; we adjust the corresponding weight upwards during training.<br />
<br />
These weights are determined through the dynamic routing procedure:<br />
[[File:Routing Algo.png|900px]]<br />
<br />
<br />
Although dynamic routing is not the only manner in which we can encode relationships between capsules, the premise of the paper is to demonstrate the capabilities of capsules under a simple implementation. Since the paper's release in 2017, numerous alternative routing implementations have been released including an EM matrix routing algorithm by the same authors (ICLR 208).<br />
<br />
=Architecture=<br />
The capsule network architecture given by the authors has 11.36 million trainable parameters. The paper itself is not very detailed on exact implementation of each architectural layer, and hence it leaves some degree of ambiguity on coding various aspects of the original network. The capsule network has 6 overall layers, with the first three layers denoting components of the encoder, and the last 3 denoting components of the decoder.<br />
<br />
==Loss Function==<br />
[[File:Loss Function.png|900px]]<br />
<br />
The cost function looks very complicated, but can be broken down into intuitive components. Before diving into the equation, remember that the length of the vector denotes the probability of object existence. The left side of the equation denotes loss when the network classifies an observation correctly; the term becomes zero when classification is incorrect. To compute loss when the network correctly classifies the label, we subtract the vector norm from a fixed quantity <math>m^+ := 0.9</math>. On the other hand, when the network classifies a label incorrectly, we penalize the loss based on the network's confidence in the incorrect label; we compute the loss by subtracting <math>m^- := 0.1</math> from the vector norm.<br />
<br />
A graphical representation of loss function values under varying vector norms is given below.<br />
[[File:Loss function chart.png|900px]]<br />
<br />
==Encoder Layers==<br />
All experiments within this paper were conducted on the MNIST dataset, and thus the architecture is built to classify the corresponding dataset. For more complex datasets, the experiments were less promising. <br />
<br />
[[File:Architecture.png|center|900px]]<br />
<br />
The encoder layer takes in a 28x28 MNIST image, and learns a 16 dimensional representation of instantiation parameters.<br />
<br />
'''Layer 1: Convolution''': <br />
This layer is a standard convolution layer. Using kernels with size 9x9x1, a stride of 1, and a ReLU activation function, we detect the 2D features within the network.<br />
<br />
'''Layer 2: PrimaryCaps''': <br />
We represent the low level features detected during convolution as 32 primary capsules. Each capsule applies eight convolutional kernels with stride 2 to the output of the convolution layer, and feeds the corresponding transformed tensors into the DigiCaps layer.<br />
<br />
'''Layer 3: DigiCaps''': <br />
This layer contains 10 digit capsules, one for each digit. As explained in the dynamic routing procedure, each input vector from the PrimaryCaps layer has its own corresponding weight matrix <math>W_{ij}</math>. Using the routing coefficients <math>c_{ij}</math> and temporary coefficients <math>b_{ij}</math>, we train the DigiCaps layer to output a ten 16 dimensional vectors. The length of the <math>i^{th}</math> vector in this layer corresponds to the probability of detection of digit <math>i</math>.<br />
<br />
==Decoder Layers==<br />
The decoder layer aims to train the capsules to extract meaningful features for image detection/classification. During training, it takes the 16 layer instantiation vector of the correct (not predicted) DigiCaps layer, and attempts to recreate the 28x28 MNIST image as best as possible. Setting the loss function as reconstruction error (Euclidean distance between reconstructed image and original image), we tune the capsules to encode features that are meaningful within the actual image.<br />
<br />
[[File:Decoder.png|center|900px]]<br />
<br />
The layer consists of three fully connected layers, and transforms a 16x1 vector from the encoder layer into a 28x28 image.<br />
<br />
In addition to the digicaps loss function, we add reconstruction error as a form of regularization. We minimize the Euclidean distance between the outputs of the logistic units and the pixel intensities of the original and reconstructed images. We scale down this reconstruction loss by 0.0005 so that it does not dominate the margin loss during training. As illustrated below, reconstructions from the 16D output of the CapsNet are robust while keeping only important details.<br />
<br />
[[File:Reconstruction.png|center|900px]]<br />
<br />
=MNIST Experimental Results=<br />
<br />
==Accuracy==<br />
The paper tests on the MNIST dataset with 60K training examples, and 10K testing. Wan et al. [2013] achieves 0.21% test error with ensembling and augmenting the data with rotation and scaling. They achieve 0.39% without them. As shown in Table 1, the authors manage to achieve 0.25% test error with only a 3 layer network; the previous state of the art only beat this number with very deep networks. This example shows the importance of routing and reconstruction regularizer, which boosts the performance. On the other hand, while the accuracies are very high, the number of parameters is much smaller compared to the baseline model.<br />
<br />
[[File:Accuracies.png|center|900px]]<br />
<br />
==What Capsules Represent for MNIST==<br />
The following figure shows the digit representation under capsules. Each row shows the reconstruction when one of the 16 dimensions in the DigitCaps representation is tweaked by intervals of 0.05 in the range [−0.25, 0.25]. By tweaking the values, we notice how the reconstruction changes, and thus get a sense for what each dimension is representing. The authors found that some dimensions represent global properties of the digits, while other represent localized properties. <br />
[[File:CapsuleReps.png|center|900px]]<br />
<br />
One example the authors provide is: different dimensions are used for the length of the ascender of a 6 and the size of the loop. The variations include stroke thickness, skew and width, as well as digit-specific variations. The authors are able to show dimension representations using a decoder network by feeding a perturbed vector.<br />
<br />
==Robustness of CapsNet==<br />
The authors conclude that DigitCaps capsules learn more robust representations for each digit class than traditional CNNs. The trained CapsNet becomes moderately robust to small affine transformations in the test data.<br />
<br />
To compare the robustness of CapsNet to affine transformations against traditional CNNs, both models (CapsNet and a traditional CNN with MaxPooling and DropOut) were trained on a padded and translated MNIST training set, in which each example is an MNIST digit placed randomly on a black background of 40 × 40 pixels. The networks were then tested on the affNIST4 dataset (MNIST digits with random affine transformation). An under-trained CapsNet which achieved 99.23% accuracy on the MNIST test set achieved a corresponding 79% accuracy on the affnist test set. A traditional CNN achieved similar accuracy (99.22%) on the mnist test set, but only 66% on the affnist test set.<br />
<br />
=MultiMNIST & Other Experiments=<br />
<br />
==MultiMNIST==<br />
To evaluate the performance of the model on highly overlapping digits, the authors generate a 'MultiMNIST' dataset. In MultiMNIST, images are two overlaid MNIST digits of the same set(train or test) but different classes. The results indicate a classification error rate of 5%. Additionally, CapsNet can be used to segment the image into the two digits that compose it. <br />
<br />
There are some additional steps to generating the MultiMNIST dataset.<br />
<br />
1. Both images are shifted by up to 4 pixels in each direction. Bounding boxes of digits in MNIST overlap by approximately 80%, so this is used to make both digits identifiable (since there is no RGB difference learnable by the network to separate the digits)<br />
<br />
2. The label becomes a vector of two numbers, representing the original digit and the randomly generated (and overlaid) digit.<br />
<br />
<br />
[[File:CapsuleNets MultiMNIST.PNG|center|700px]]<br />
<br />
=Critique=<br />
Although the network performs incredibly favourably in the author's experiments, it has a long way to go on more complex datasets. On CIFAR 10, the network achieved subpar results, and the experimental results seem to worsen the more complex the problem becomes. This is to be expected, since these networks are still in their early stage; later innovations might come in the upcoming decades/years.<br />
<br />
Hinton talks about CapsuleNets revolutionizing areas such as self-driving, but such groundbreaking innovations are a long ways away from CIFAR10, and even further from MNIST. Only will time tell if CapsNets will live up to their hype.<br />
<br />
Capsules inherently segment images, and learn a lower dimensional embedding in a new manner, which makes them likely to perform well on segmentation and computer vision tasks once further research is done on them. <br />
<br />
Additionally these networks are more interpretable than CNNs, and have strong theoretical reasoning for why they could work. Naturally, it would be hard for a new architecture to beat the heavily researched/modified CNNs.<br />
<br />
* ([https://openreview.net/forum?id=HJWLfGWRb]) it's not fully clear how effective it can be performed / how scalable it is. Evaluation is performed on a small dataset for shape recognition. The approach will need to be tested on larger, more challenging datasets.<br />
<br />
=Future Work=<br />
The same authors [N. F. Geoffrey E Hinton, Sara Sabour] presented another paper "MATRIX CAPSULES WITH EM ROUTING" in ICLR 2018, which achieved better results than the work presented in this paper. They presented a novel capsule type, where each capsule has a logistic unit and a 4x4 pose matrix. This new type reduced number of errors by 45%, and performed better than standard CNN on white box adversarial attacks.<br />
<br />
=References=<br />
#N. F. Geoffrey E Hinton, Sara Sabour. Matrix capsules with em routing. In International Conference on Learning Representations, 2018.<br />
#S. Sabour, N. Frosst, and G. E. Hinton, “Dynamic routing between capsules,” arXiv preprint arXiv:1710.09829v2, 2017<br />
# Hinton, G. E., Krizhevsky, A. and Wang, S. D. (2011), Transforming Auto-encoders <br />
#Geoffrey Hinton's talk: What is wrong with convolutional neural nets? - Talk given at MIT. Brain & Cognitive Sciences - Fall Colloquium Series. [https://www.youtube.com/watch?v=rTawFwUvnLE ]<br />
#Understanding Hinton’s Capsule Networks - Max Pechyonkin's series [https://medium.com/ai%C2%B3-theory-practice-business/understanding-hintons-capsule-networks-part-i-intuition-b4b559d1159b]</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=CapsuleNets&diff=41029CapsuleNets2018-11-22T22:58:35Z<p>Gchalato: /* MultiMNIST */</p>
<hr />
<div>The paper "Dynamic Routing Between Capsules" was written by three researchers at Google Brain: Sara Sabour, Nicholas Frosst, and Geoffrey E. Hinton. This paper was published and presented at the 31st Conference on Neural Information Processing Systems (NIPS 2017) in Long Beach, California. The same three researchers recently published a highly related paper "Matrix Capsules with EM Routing" for ICLR 2018.<br />
<br />
=Motivation=<br />
<br />
Ever since AlexNet eclipsed the performance of competing architectures in the 2012 ImageNet challenge, convolutional neural networks have maintained their dominance in computer vision applications. Despite the recent successes and innovations brought about by convolutional neural networks, some assumptions made in these networks are perhaps unwarranted and deficient. Using a novel neural network architecture, the authors create CapsuleNets, a network that they claim is able to learn image representations in a more robust, human-like manner. With only a 3 layer capsule network, they achieved near state-of-the-art results on MNIST.<br />
==Adversarial Examples==<br />
<br />
First discussed by Christian Szegedy et. al. in late 2013, adversarial examples have been heavily discussed by the deep learning community as a potential security threat to AI learning. Adversarial examples are defined as inputs that an attacker creates intentionally fool a machine learning model. An example of an adversarial example is shown below: <br />
<br />
[[File:adversarial_img_1.png |center]]<br />
To the human eye, the image appears to be a panda both before and after noise is injected into the image, whereas the trained ConvNet model discerns the noisy image as a Gibbon with almost 100% certainty. The fact that the network is unable to classify the above image as a panda after the epsilon perturbation leads to many potential security risks in AI dependent systems such as self-driving vehicles. Although various methods have been suggested to combat adversarial examples, robust defences are hard to construct due to the inherent difficulties in constructing theoretical models for the adversarial example crafting process. However, beyond the fact that these examples may serve as a security threat, it emphasizes that these convolutional neural networks do not learn image classification/object detection patterns the same way that a human would. Rather than identifying the core features of a panda such as: its eyes, mouth, nose, and the gradient changes in its black/white fur, the convolutional neural network seems to be learning image representations in a completely different manner. Deep learning researchers often attempt to model neural networks after human learning, and it is clear that further steps must be taken to robustify ConvNets against targeted noise perturbations.<br />
<br />
==Drawbacks of CNNs==<br />
Hinton claims that the key fault with traditional CNNs lies within the pooling function. Although pooling builds translational invariance into the network, it fails to preserve spatial relationships between objects. When we pool, we effectively reduce a kxk kernel of convolved cells into a scalar input. This results in a desired local invariance without inhibiting the network's ability to detect features, but causes valuable spatial information to be lost.<br />
<br />
In the example below, the network is able to detect the similar features (eyes, mouth, nose, etc) within both images, but fails to recognize that one image is a human face, while the other is a Picasso-esque due to the CNN's inability to encode spatial relationships after multiple pooling layers.<br />
<br />
<br />
[[File:Equivariance Face.png |center]]<br />
<br />
Conversely, we hope that a CNN can recognize that both of the following pictures contain a kitten. Unfortunately, when we feed the two images into a ResNet50 architecture, only the first image is correctly classified, while the second image is predicted to be a guinea pig.<br />
<br />
<br />
[[File:kitten.jpeg |center]]<br />
<br />
<br />
[[File:kitten-rotated-180.jpg |center]]<br />
<br />
For a more in depth discussion on the problems with ConvNets, please listen to Geoffrey Hinton's talk "What is wrong with convolutional neural nets?" given at MIT during the Brain & Cognitive Sciences - Fall Colloquium Series (December 4, 2014).<br />
<br />
==Intuition for Capsules==<br />
Human vision ignores irrelevant details by using a carefully determined sequence of fixation points to ensure that only a tiny fraction of the optic array is ever processed at the highest resolution. Hinton argues that our brains reason visual information by deconstructing it into a hierarchical representation which we then match to familiar patterns and relationships from memory. The key difference between this understanding and the functionality of CNNs is that recognition of an object should not depend on the angle from which it is viewed. <br />
<br />
To enforce rotational and translational equivariance, Capsule Networks store and preserve hierarchical pose relationships between objects. The core idea behind capsule theory is the explicit numerical representations of relative relationships between different objects within an image. Building these relationships into the Capsule Networks model, the network is able to recognize newly seen objects as a rotated view of a previously seen object. For example, the below image shows the Statue of Liberty under five different angles. If a person had only seen the Statue of Liberty from one angle, they would be able to ascertain that all five pictures below contain the same object (just from a different angle).<br />
<br />
[[File:Rotational Invariance.jpeg |center]]<br />
<br />
Building on this idea of hierarchical representation of spatial relationships between key entities within an image, the authors introduce Capsule Networks. Unlike traditional CNNs, Capsule Networks are better equipped to classify correctly under rotational invariance. Furthermore, the authors managed to achieve state of the art results on MNIST using a fraction of the training samples that alternative state of the art networks require.<br />
<br />
<br />
=Background, Notation, and Definitions=<br />
<br />
==What is a Capsule==<br />
"Each capsule learns to recognize an implicitly defined visual entity over a limited domain of viewing conditions and deformations and it outputs both the probability that the entity is present within its limited domain and a set of “instantiation parameters” that may include the precise pose, lighting and deformation of the visual entity relative to an implicitly defined canonical version of that entity. When the capsule is working properly, the probability of the visual entity being present is locally invariant — it does not change as the entity moves over the manifold of possible appearances within the limited domain covered by the capsule. The instantiation parameters, however, are “equivariant” — as the viewing conditions change and the entity moves over the appearance manifold, the instantiation parameters change by a corresponding amount because they are representing the intrinsic coordinates of the entity on the appearance manifold."<br />
<br />
In essence, capsules store object properties in a vector form; probability of detection is encoded as the vector's length, while spatial properties are encoded as the individual vector components. Thus, when a feature is present but the image captures it under a different angle, the probability of detection remains unchanged.<br />
<br />
==Notation==<br />
<br />
We want the length of the output vector of a capsule to represent the probability that the entity represented by the capsule is present in the current input. The paper performs a non-linear squashing operation to ensure that vector length falls between 0 and 1, with shorter vectors (less likely to exist entities) being shrunk towards 0. <br />
<br />
\begin{align} \mathbf{v}_j &= \frac{||\mathbf{s}_j||^2}{1+ ||\mathbf{s}_j||^2} \frac{\mathbf{s}_j}{||\mathbf{s}_j||^2} \end{align}<br />
<br />
where <math>\mathbf{v}_j</math> is the vector output of capsule <math>j</math> and <math>s_j</math> is its total input.<br />
<br />
For all but the first layer of capsules, the total input to a capsule <math>s_j</math> is a weighted sum over all “prediction vectors” <math>\hat{\mathbf{u}}_{j|i}</math> from the capsules in the layer below and is produced by multiplying the output <math>\mathbf{u}i</math> of a capsule in the layer below by a weight matrix <math>\mathbf{W}ij</math><br />
<br />
\begin{align}<br />
\mathbf{s}_j = \sum_i c_{ij}\hat{\mathbf{u}}_{j|i}, \hat{\mathbf{u}}_{j|i}= \mathbf{W}_{ij}\mathbf{u}_i<br />
\end{align}<br />
where the <math>c_{ij}</math> are coupling coefficients that are determined by the iterative dynamic routing process.<br />
<br />
The coupling coefficients between capsule <math>i</math> and all the capsules in the layer above sum to 1 and are determined by a “routing softmax” whose initial logits <math>b_{ij}</math> are the log prior probabilities that capsule <math>i</math> should be coupled to capsule <math>j</math>.<br />
<br />
\begin{align}<br />
c_{ij} = \frac{\exp(b_{ij})}{\sum_k \exp(b_{ij})}<br />
\end{align}<br />
=Network Training and Dynamic Routing=<br />
<br />
==Understanding Capsules==<br />
The notation can get somewhat confusing, so I will provide intuition behind the computational steps within a capsule. The following image is taken from naturomic's talk on Capsule Networks.<br />
<br />
[[File:CapsuleNets.jpeg|center|800px]]<br />
<br />
The above image illustrates the key mathematical operations happening within a capsule (and compares them to the structure of a neuron). Although the operations are rather straightforward, it's crucial to note that the capsule computes an affine transformation onto each input vector. The length of the input vectors <math>\mathbf{u}_{i}</math> represent the probability of entity <math>i</math> existing in a lower level. This vector is then reoriented with an affine transform using <math>\mathbf{W}_{ij}</math> matrices that encode spatial relationships between entity <math>\mathbf{u}_{i}</math> and other lower level features.<br />
<br />
We illustrate the intuition behind vector-to-vector matrix multiplication within capsules using the following example: if vectors <math>\mathbf{u}_{1}</math>, <math>\mathbf{u}_{2}</math>, and <math>\mathbf{u}_{3}</math> represent detection of eyes, nose, and mouth respectively, then after multiplication with trained weight matrices <math>\mathbf{W}_{ij}</math> (where j denotes existence of a face), we should get a general idea of the general location of the higher level feature (face), similar to the image below.<br />
<br />
[[File:Predictions.jpeg |center]]<br />
<br />
==Dynamic Routing==<br />
A capsule <math>i</math> in a lower-level layer needs to decide how to send its output vector to higher-level capsules <math>j</math>. This decision is made with probability proportional to <math>c_{ij}</math>. If there are <math>K</math> capsules in the level that capsule <math>i</math> routes to, then we know the following properties about <math>c_{ij}</math>: <math>\sum_{j=1}^M c_{ij} = 1, c_{ij} \geq 0</math><br />
<br />
In essence, the <math>\{c_{ij}\}_{j=1}^M</math> denotes a discrete probability distribution with respect to capsule <math>i</math>'s output location. Lower level capsules decide which higher level capsules to send vectors into by adjusting the corresponding routing weights <math>\{c_{ij}\}_{j=1}^M</math>. After a few iterations in training, numerous vectors will have already been sent to all higher level capsules. Based on the similarity between the current vector being routed and all vectors already sent into the higher level capsules, we decide which capsule to send the current vector into.<br />
[[File:Dynamic Routing.png|center|900px]]<br />
<br />
In the image above, we notice that a cluster of points similar to the current vector has already been routed into capsule K, while most points in capsule J are high dissimilar. It thus makes more sense to route the current observation into capsule K; we adjust the corresponding weight upwards during training.<br />
<br />
These weights are determined through the dynamic routing procedure:<br />
[[File:Routing Algo.png|900px]]<br />
<br />
<br />
Although dynamic routing is not the only manner in which we can encode relationships between capsules, the premise of the paper is to demonstrate the capabilities of capsules under a simple implementation. Since the paper's release in 2017, numerous alternative routing implementations have been released including an EM matrix routing algorithm by the same authors (ICLR 208).<br />
<br />
=Architecture=<br />
The capsule network architecture given by the authors has 11.36 million trainable parameters. The paper itself is not very detailed on exact implementation of each architectural layer, and hence it leaves some degree of ambiguity on coding various aspects of the original network. The capsule network has 6 overall layers, with the first three layers denoting components of the encoder, and the last 3 denoting components of the decoder.<br />
<br />
==Loss Function==<br />
[[File:Loss Function.png|900px]]<br />
<br />
The cost function looks very complicated, but can be broken down into intuitive components. Before diving into the equation, remember that the length of the vector denotes the probability of object existence. The left side of the equation denotes loss when the network classifies an observation correctly; the term becomes zero when classification is incorrect. To compute loss when the network correctly classifies the label, we subtract the vector norm from a fixed quantity <math>m^+ := 0.9</math>. On the other hand, when the network classifies a label incorrectly, we penalize the loss based on the network's confidence in the incorrect label; we compute the loss by subtracting <math>m^- := 0.1</math> from the vector norm.<br />
<br />
A graphical representation of loss function values under varying vector norms is given below.<br />
[[File:Loss function chart.png|900px]]<br />
<br />
==Encoder Layers==<br />
All experiments within this paper were conducted on the MNIST dataset, and thus the architecture is built to classify the corresponding dataset. For more complex datasets, the experiments were less promising. <br />
<br />
[[File:Architecture.png|center|900px]]<br />
<br />
The encoder layer takes in a 28x28 MNIST image, and learns a 16 dimensional representation of instantiation parameters.<br />
<br />
'''Layer 1: Convolution''': <br />
This layer is a standard convolution layer. Using kernels with size 9x9x1, a stride of 1, and a ReLU activation function, we detect the 2D features within the network.<br />
<br />
'''Layer 2: PrimaryCaps''': <br />
We represent the low level features detected during convolution as 32 primary capsules. Each capsule applies eight convolutional kernels with stride 2 to the output of the convolution layer, and feeds the corresponding transformed tensors into the DigiCaps layer.<br />
<br />
'''Layer 3: DigiCaps''': <br />
This layer contains 10 digit capsules, one for each digit. As explained in the dynamic routing procedure, each input vector from the PrimaryCaps layer has its own corresponding weight matrix <math>W_{ij}</math>. Using the routing coefficients <math>c_{ij}</math> and temporary coefficients <math>b_{ij}</math>, we train the DigiCaps layer to output a ten 16 dimensional vectors. The length of the <math>i^{th}</math> vector in this layer corresponds to the probability of detection of digit <math>i</math>.<br />
<br />
==Decoder Layers==<br />
The decoder layer aims to train the capsules to extract meaningful features for image detection/classification. During training, it takes the 16 layer instantiation vector of the correct (not predicted) DigiCaps layer, and attempts to recreate the 28x28 MNIST image as best as possible. Setting the loss function as reconstruction error (Euclidean distance between reconstructed image and original image), we tune the capsules to encode features that are meaningful within the actual image.<br />
<br />
[[File:Decoder.png|center|900px]]<br />
<br />
The layer consists of three fully connected layers, and transforms a 16x1 vector from the encoder layer into a 28x28 image.<br />
<br />
In addition to the digicaps loss function, we add reconstruction error as a form of regularization. We minimize the Euclidean distance between the outputs of the logistic units and the pixel intensities of the original and reconstructed images. We scale down this reconstruction loss by 0.0005 so that it does not dominate the margin loss during training. As illustrated below, reconstructions from the 16D output of the CapsNet are robust while keeping only important details.<br />
<br />
[[File:Reconstruction.png|center|900px]]<br />
<br />
=MNIST Experimental Results=<br />
<br />
==Accuracy==<br />
The paper tests on the MNIST dataset with 60K training examples, and 10K testing. Wan et al. [2013] achieves 0.21% test error with ensembling and augmenting the data with rotation and scaling. They achieve 0.39% without them. As shown in Table 1, the authors manage to achieve 0.25% test error with only a 3 layer network; the previous state of the art only beat this number with very deep networks. This example shows the importance of routing and reconstruction regularizer, which boosts the performance. On the other hand, while the accuracies are very high, the number of parameters is much smaller compared to the baseline model.<br />
<br />
[[File:Accuracies.png|center|900px]]<br />
<br />
==What Capsules Represent for MNIST==<br />
The following figure shows the digit representation under capsules. Each row shows the reconstruction when one of the 16 dimensions in the DigitCaps representation is tweaked by intervals of 0.05 in the range [−0.25, 0.25]. By tweaking the values, we notice how the reconstruction changes, and thus get a sense for what each dimension is representing. The authors found that some dimensions represent global properties of the digits, while other represent localized properties. <br />
[[File:CapsuleReps.png|center|900px]]<br />
<br />
One example the authors provide is: different dimensions are used for the length of the ascender of a 6 and the size of the loop. The variations include stroke thickness, skew and width, as well as digit-specific variations. The authors are able to show dimension representations using a decoder network by feeding a perturbed vector.<br />
<br />
==Robustness of CapsNet==<br />
The authors conclude that DigitCaps capsules learn more robust representations for each digit class than traditional CNNs. The trained CapsNet becomes moderately robust to small affine transformations in the test data.<br />
<br />
To compare the robustness of CapsNet to affine transformations against traditional CNNs, both models (CapsNet and a traditional CNN with MaxPooling and DropOut) were trained on a padded and translated MNIST training set, in which each example is an MNIST digit placed randomly on a black background of 40 × 40 pixels. The networks were then tested on the affNIST4 dataset (MNIST digits with random affine transformation). An under-trained CapsNet which achieved 99.23% accuracy on the MNIST test set achieved a corresponding 79% accuracy on the affnist test set. A traditional CNN achieved similar accuracy (99.22%) on the mnist test set, but only 66% on the affnist test set.<br />
<br />
=MultiMNIST & Other Experiments=<br />
<br />
==MultiMNIST==<br />
To evaluate the performance of the model on highly overlapping digits, the authors generate a 'MultiMNIST' dataset. In MultiMNIST, images are two overlaid MNIST digits of the same set(train or test) but different classes. The results indicate a classification error rate of 5%. Additionally, CapsNet can be used to segment the image into the two digits that compose it. <br />
<br />
There are a number of additional steps to generating the MultiMNIST dataset.<br />
<br />
1. Both images are shifted by up to 4 pixels in each direction. Bounding boxes of digits in MNIST overlap by approximately 80%, so this is used to make both digits identifiable (since there is no RGB difference learnable by the network to separate the digits)<br />
<br />
2. The label becomes a vector of two numbers, representing the original digit and the randomly generated (and overlaid) digit.<br />
<br />
<br />
[[File:CapsuleNets MultiMNIST.PNG|center|700px]]<br />
<br />
=Critique=<br />
Although the network performs incredibly favourably in the author's experiments, it has a long way to go on more complex datasets. On CIFAR 10, the network achieved subpar results, and the experimental results seem to worsen the more complex the problem becomes. This is to be expected, since these networks are still in their early stage; later innovations might come in the upcoming decades/years.<br />
<br />
Hinton talks about CapsuleNets revolutionizing areas such as self-driving, but such groundbreaking innovations are a long ways away from CIFAR10, and even further from MNIST. Only will time tell if CapsNets will live up to their hype.<br />
<br />
Capsules inherently segment images, and learn a lower dimensional embedding in a new manner, which makes them likely to perform well on segmentation and computer vision tasks once further research is done on them. <br />
<br />
Additionally these networks are more interpretable than CNNs, and have strong theoretical reasoning for why they could work. Naturally, it would be hard for a new architecture to beat the heavily researched/modified CNNs.<br />
<br />
* ([https://openreview.net/forum?id=HJWLfGWRb]) it's not fully clear how effective it can be performed / how scalable it is. Evaluation is performed on a small dataset for shape recognition. The approach will need to be tested on larger, more challenging datasets.<br />
<br />
=Future Work=<br />
The same authors [N. F. Geoffrey E Hinton, Sara Sabour] presented another paper "MATRIX CAPSULES WITH EM ROUTING" in ICLR 2018, which achieved better results than the work presented in this paper. They presented a novel capsule type, where each capsule has a logistic unit and a 4x4 pose matrix. This new type reduced number of errors by 45%, and performed better than standard CNN on white box adversarial attacks.<br />
<br />
=References=<br />
#N. F. Geoffrey E Hinton, Sara Sabour. Matrix capsules with em routing. In International Conference on Learning Representations, 2018.<br />
#S. Sabour, N. Frosst, and G. E. Hinton, “Dynamic routing between capsules,” arXiv preprint arXiv:1710.09829v2, 2017<br />
# Hinton, G. E., Krizhevsky, A. and Wang, S. D. (2011), Transforming Auto-encoders <br />
#Geoffrey Hinton's talk: What is wrong with convolutional neural nets? - Talk given at MIT. Brain & Cognitive Sciences - Fall Colloquium Series. [https://www.youtube.com/watch?v=rTawFwUvnLE ]<br />
#Understanding Hinton’s Capsule Networks - Max Pechyonkin's series [https://medium.com/ai%C2%B3-theory-practice-business/understanding-hintons-capsule-networks-part-i-intuition-b4b559d1159b]</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Annotating_Object_Instances_with_a_Polygon_RNN&diff=40812Annotating Object Instances with a Polygon RNN2018-11-22T06:05:13Z<p>Gchalato: /* Related Works */</p>
<hr />
<div>Summary of the CVPR '17 best [https://www.cs.utoronto.ca/~fidler/papers/paper_polyrnn.pdf ''paper'']<br />
<br />
The presentation video of paper is available here[https://www.youtube.com/watch?v=S1UUR4FlJ84].<br />
<br />
= Background =<br />
<br />
If a snapshot of an image is given to a human, how will he/she describe a scene? He/she might identify that there is a car parked near the curb, or that the car is parked right beside a street light. This ability to decompose objects in scenes into separate entities is key to understanding what is around us and it helps to reason about the behavior of objects in the scene.<br />
<br />
Automating this process is a classic computer vision problem and is often termed "object detection". There are four distinct levels of detection (refer to Figure 1 for a visual cue):<br />
<br />
1. Classification + Localization: This is the most basic method that detects whether '''an''' object is either present or absent in the image and then identifies the position of the object within the image in the form of a bounding box overlayed on the image.<br />
<br />
2. Object Detection: The classic definition of object detection points to the detection and localization of '''multiple''' objects of interest in the image. The output of the detection is still a bounding box overlayed on the image at the position corresponding to the location of the objects in the image.<br />
<br />
3. Semantic Segmentation: This is a pixel level approach, i.e., each pixel in the image is assigned to a category label. Here, there is no difference between instances; this is to say that there are objects present from three distinct categories in the image, without tracking or reporting the number of appearances of each instance within a category. <br />
<br />
4. Instance Segmentation (''This paper performs this''): The goal is to not only to assign pixel-level categorical labels, but to identify each entity separately as sheep 1, sheep 2, sheep 3, grass, and so on.<br />
<br />
[[File:Figure_1.jpeg | 450px|thumb|center|Figure 1: Different levels of detection in an image.]]<br />
<br />
<br />
== Motivation ==<br />
<br />
Semantic segmentation helps us achieve a deeper understanding of images than image classification or object detection. Over and above this, instance segmentation is crucial in applications where multiple objects of the same category are to be tracked, especially in autonomous driving, mobile robotics, and medical image processing. This paper deals with a novel method to tackle the instance segmentation problem pertaining specifically to the field of autonomous driving, but shown to generalize well in other fields such as medical image processing.<br />
A polygon is natural form of annotation. Current instant segmentations annotated by humans use polygons because it is a special representation of the image which can use small number of vertices instead of various pixels and makes it easy to incorporate user modifications.<br />
<br />
[[File:polygon.png|600px|center]]<br />
<br />
== Goal ==<br />
<br />
Most of the recent approaches to on instance segmentation are based on deep neural networks and have demonstrated impressive performance. Given that these approaches require a lot of computational resources and that their performance depends on the amount of accessible training data, there has been an increase in the demand to label/annotate large-scale datasets. This is both expensive and time-consuming. <br />
<br />
{| class=wikitable width=700 align=center<br />
|Thus, the '''main goal''' of the paper is to enable '''semi-automatic''' annotation of object instances.<br />
|}<br />
<br />
Figure 2 demonstrates how the interface looks like for better clarity.<br />
<br />
Most of the datasets available pass through a stage where annotators manually outline the objects with a closed polygon. Polygons allow annotation of objects with a small number of clicks (30 - 40) compared to other methods. This approach works as the silhouette of an object is typically connected without holes. <br />
<br />
{| class=wikitable width=900 align=center<br />
|Thus, the authors suggest to adopt this same technique to annotate images using polygons, except they plan to automate the method and replace/reduce manual labeling. The '''intuition''' behind the success of this method is the '''sparse''' nature of these polygons that allow annotating of an object through a cluster of pixels rather than classification at the pixel-level.<br />
|}<br />
<br />
[[File:Annotating Object Instances Example.png | 450px|thumb|center|Figure 2: Given a bounding box, polygon outlining the the object instance inside the box is predicted. This approach is designed to facilitation annotation, and easily incorporates user corrections of points to improve the overall object’s polygon. ]]<br />
<br />
<br />
= Related Works =<br />
<br />
Some of the techniques used in semi-automatic annotation are as follows:<br />
<br />
1. '''GrabCut''': In general, GrabCut is a method to separate the foreground and background of an image with minimal user interaction. Specifically, the user need only create a rectangular bounding box containing the foreground, and the algorithm will extract the object in the foreground. A major contribution of the paper is that labelling (of the object in the foreground) was not required, as the algorithm was able to identify where significant changes in colour pattern occurred. In this sense, it mimics automatic segmentation when combined with a Region Proposal Network. <br />
<br />
[[File:GrabCut_Example.png | 450px|thumb|center|Figure 3: Illustration of GrabCut.]]<br />
<br />
2. '''GrabCut + CNN''': Scribbles have also been used to train CNNs for semantic image segmentation. <br />
<br />
3. '''Superpixels''': Superpixels in the form of small polygons where the color intensity within each superpixel is similar, to a certain threshold, have been used to provide a sparse representation of the large number of pixels in an image. However, the performance of this technique depends on the scale of the superpixels and hence sometimes merges small objects.<br />
<br />
[[File:Superpixel_idea.jpg | 450px|thumb|center|Figure 4: Illustration of the superpixel idea.]]<br />
<br />
= Model =<br />
<br />
As an '''input''' to the model, an annotator or perhaps another neural network provides a bounding box containing an object of interest and the model auto-generates a polygon outlining the object instance using a Recurrent Neural Network which they call: Polygon-RNN.<br />
<br />
The RNN model predicts the vertices of the polygon at each time step given a CNN representation of the image, the last two time steps, and the first vertex location. The location of the first vertex is defined differently and will be defined shortly. The information regarding the previous two-time steps helps the RNN create a polygon in a specific direction and the first vertex provides a cue for loop closure of the polygon edges.<br />
<br />
The polygon is parametrized as a sequence of 2D vertices and it is assumed that the polygon is closed. In addition, the polygon generation is fixed to follow a clockwise orientation since there are multiple ways to create a polygon given that it is cyclic structure. However, the starting point of the sequence is defined so that it can be any of the vertices of the polygon.<br />
<br />
== Architecture ==<br />
<br />
There are two primary networks at play: 1. CNN with skip connections, and 2. One-to-many type RNN.<br />
<br />
[[File:Figure_2_Neel.JPG | 800px|thumb|center|Figure 5: Model architecture for Polygon-RNN depicting a CNN with skip connections feeding into a 2 layer ConvLSTM (One-to-many type) ('''Note''': A possible point of confusion - the authors have only shown the layers of VGG16 architecture here that have the skip connections introduced).]]<br />
<br />
1. '''CNN with skip connections''':<br />
<br />
The authors have adopted the VGG16 feature extractor architecture with a few modifications pertaining to the preservation of features fused together in a tensor that can feed into the RNN (refer to Figure 5). Namely, the last max-pooling layer (''pool5'') present in the VGG16 CNN has been removed. The image fed into the CNN is pre-shrunk to a 224x224x3 tensor(3 being the Red, Green, and Blue channels). The image passes through 2 pooling layers and 2 convolutional layers. Since, the features extracted after each operation are to be preserved and fused later on, at each of these four steps, the idea is to have a tensor with a common width of 512; so the output tensor at pool2 is convolved with 4 3x3x128 filters and the output tensor at pool3 is convolved with 2 3x3x256 filters. The skip connections from the four layers allow the CNN to extract low-level edge and corner features (helps to follow the object's boundaries) as well as boundary/semantic information about the instances (helps to identify the object). Finally, a 3x3 convolution applied along with a ReLU non-linearity results in a 28x28x128 tensor that contains semantic information pertinent to the image frame and is taken as an input by the RNN.<br />
<br />
2. '''RNN - 2 Layer ConvLSTM'''<br />
<br />
The RNN is employed to capture information about the previous vertices in the time-series. Specifically, a Convolutional LSTM is used as a decoder. The ConvLSTM allows preservation of the spatial information in 2D received from CNN and reduces the number of parameters compared to a Fully Connected RNN. The polygon is modeled with a kernel size of 3x3 and 16 channels outputting a vertex at each time step. The ConvLSTM gets as input a tensor step t which<br />
concatenates 4 features: the CNN feature representation of the image, one-hot encoding of the previous predicted vertex and the vertex predicted<br />
from two time steps ago, as well as the one-hot encoding of the first predicted vertex. <br />
<br />
The Convolutional LSTM computes the hidden state <math display = "inline">h_t</math> given the input <math display = "inline">x_t</math> based on the following equations:<br />
<center><br />
<math display="block"><br />
\begin{pmatrix}<br />
i_t \\<br />
f_t \\<br />
o_t \\<br />
g_t \\<br />
\end{pmatrix}<br />
= W_h * h_{t-1} + W_x * x_t + b<br />
</math><br />
<br />
<math display="block"><br />
c_t = \sigma(f_t) \bigodot c_{t-1} + \sigma(i_t) \bigodot tanh(g_t)<br />
</math><br />
<br />
<math display="block"><br />
h_t = \sigma(o_t) \bigodot tanh(c_t)<br />
</math><br />
</center><br />
where <math display = "inline">i, f, o</math> denote the input, forget, and output gate, <math display = "inline">h</math> is the hidden state and <math display = "inline">c</math> is the cell state. Also, <math display = "inline">\sigma</math> denotes the sigmoid function, <math display = "inline">\bigodot</math> indicates an element-wise product and <math display = "inline">*</math> a convolution. <math display = "inline">W_h</math> denotes the hidden-to-state convolution kernel and <math display = "inline">W_x</math> the input-to-state convolution kernel.<br />
<br />
The authors have treated the vertex prediction task as a classification task in that the location of the vertices is through a one-hot representation of dimension DxD + 1 (D chosen to be 28 by the authors in tests). The one additional dimension is the storage cue for loop closure for the polygon. Given that, the one-hot representation of the two previously predicted vertices and the first vertex are taken in as an input, a clockwise (or for that reason any fixed direction) direction can be forced for the creation of the polygon. Coming back to the prediction of the first vertex, as polygon is a circle, any vertex of a polygon can be used as a starting point. Therefore the authors treat the starting point as special, and this is done through further modification of the CNN by adding two DxD layers with one branch predicting object instance boundaries while the other takes in this output as well as the image features to predict vertices of the polygon. The boundaries and vertices prediction are being treated as binary classification problem in each cell in the output grid. This CNN is trained separately. Here, <math display = "inline">y_t</math> denotes the one-hot encoding of the vertex and is the output at time step t.<br />
<br />
== Training ==<br />
<br />
The training of the model is done as follows:<br />
<br />
1. Cross-entropy is used for the RNN loss function. To avoid over-penalizing of mispredictions, non-zero probability mass are assigned to locations which are within a distance of 2 in D × D output grid.<br />
<br />
2. Instead of Stochastic Gradient Descent, Adam is used for optimization: batch size = 8, learning rate = 1e^-4 (learning rate decays after 10 epochs by a factor of 10) <br />
<br />
3. For the first vertex prediction, the modified CNN mentioned previously, is trained using a multi-task cost function.<br />
<br />
The reported time for training is one day on a Nvidia Titan-X GPU.<br />
<br />
The resolution of the polygon is 28 x 28, based on the downsampling factor and ConvLSTM resolution. They simplified the polygon by removing vertices on the grid line and the same vertices that fall in the same grid. They also randomly flipped images, enlarged original bounding boxes and randomly selected the starting vertex of the polygon notation as their data augmentation process.<br />
<br />
== Importance of Human Annotator in the Loop ==<br />
<br />
The model allows for the prediction at a given time step to be corrected and this corrected vertex is then fed into the next time step of the RNN, effectively rejecting the network predicted vertex. This has the simple effect of putting the model "back on the right track". Note that this is only possible due to the adoption of the RNN architecture i.e. the inherent nature of the RNN to accept previous outputs allows incorporation of the user's judgement. The typical inference time as quoted by the paper is 250ms per object.<br />
<br />
= Results =<br />
<br />
== Evaluation Metrics ==<br />
<br />
The evaluation of the model performance was conducted based on the Cityscapes and KITTI Datasets. There are two metrics used for evaluation:<br />
<br />
1. '''IoU''': The standard Intersection over Union (IoU) measure is used for comparison. In add The calculation for IoU takes both the predicted and ground-truth object boundaries. The intersection (area contained in both boundaries at once) is divided by the union (the area contained by at least one, or both, of the boundaries). A low score of this metric would mean that there is little overlap between the boundaries, or large areas on non-overlap, and a score of 1.0 would indicate that the two boundaries contain the same area.<br />
<br />
2. '''Number of Clicks''': To evaluate the speed up factor, the checkerboard distance is used to measure the distance between the ground truth (GT) and the output of the Polygon RNN. A set of distance thresholds are set <math display = "inline">T &isin; [1,2,3,4]</math> and if the distance exceeds the particular threshold, the correction is made by an annotator to match the GT and the '''Number of Clicks''' is used to evaluate the speed up factor.<br />
<br />
== Baseline Techniques ==<br />
<br />
1. '''SharpMask''': a 50 layer ResNet considered as the state of the art annotation method.<br />
<br />
2. '''DeepMask''': a build-up on the 50 layer ResNet with an addition of another CNN.<br />
<br />
3. '''Dilation10''': another simple technique using purely convolutional operations.<br />
<br />
4. '''SquareBox''': a simple technique where an entire bounding box is labeled as an object<br />
<br />
== Quantitative Results ==<br />
<br />
We report the IoU metric in Table<br />
1. The Polygon RNN method outperforms the baselines in 6 out of the 8 categories and has a mean IoU greater than all of the baselines. Particularly, in the car, person, and rider categories, a 12%, 7%, and 6% higher performance than SharpMask is achieved.<br />
<br />
[[File:Table_1_Neel.JPG | 800px|thumb|center|Table 1: IoU performance on Cityscapes data without any annotator intervention.]]<br />
<br />
In addition, with the help of the annotator, the speedup factor was 7.3 times with under 5 clicks which the authors claim is the main advantage of this method.<br />
<br />
[[File:Table_0_Neel.JPG | 800px|thumb|center|Table 2: IoU performance on Cityscapes data with annotator intervention.]]<br />
<br />
The method also works well with other datasets such as KITTI:<br />
<br />
[[File:Table_2_Neel.JPG | 800px|thumb|center|Table 3: IoU performance on KITTI data.]]<br />
<br />
== Effect of object size ==<br />
In Fig. 4, we see how our model performs w.r.t baselines on different instance sizes. For small instances, our model performs significantly better than the baselines. For larger objects, the baselines have an advantage due to the larger output resolution. <br />
<br />
[[File:IoU_vs_size_of_instance.PNG | 500px|thumb|center|Fig 4: IoU_vs_size_of_instance.]]<br />
<br />
== Qualitative Results ==<br />
<br />
In addition, most of the comparisons with human annotators show that the method is at par with human-level annotation.<br />
<br />
<gallery widths=500px heights=500px perrow=2 mode="packed"><br />
File:Figure_3_Neel.JPG|Figure 6: Qualitative results: comparison with human annotator.|alt=alt language<br />
File:Figure_4_Neel.JPG|Figure 7: Qualitative results: comparison with human annotator.|alt=alt language<br />
</gallery><br />
<br />
=Conclusion=<br />
<br />
The important conclusions from this paper are:<br />
<br />
1. The paper presented a powerful generic annotation tool for modelling complex annotations as a simple polygon that works on different unseen datasets. <br />
<br />
2. Significant improvement in annotation time can be achieved with the Polygon-RNN method itself (speed-up factor of 4.74).<br />
<br />
3. However, the flexibility of having inputs from a human annotator helps increase the IoU for a certain range of clicks.<br />
<br />
4. The model architecture has a down-sampling factor of 16 and the final output resolution and accuracy is sensitive to object size.<br />
<br />
5. Another downside of the model architecture is that training time is increased due to the training of the CNN for the first vertex.<br />
<br />
=Critique=<br />
<br />
1. With the human annotator in the loop, the model speeds up the process of annotation by over 7 times which is perhaps a big cost and time cutting improvement for companies.<br />
<br />
2. Given that this model uses the VGG16 architecture compared to the 50 layer ResNet in SharpMask, this method is quite efficient.<br />
<br />
3. This paper requires training of an entire CNN for the first vertex and is inefficient in that sense as it introduces additional parameters adding to the computation time and resource demand.<br />
<br />
4. The baseline methods have an upper hand compared to this model when it comes to larger objects since the nature of the down-scaled structure adopted by this model.<br />
<br />
5. In terms of future work, elimination of the additional CNN for the first vertex as well as an enhanced architecture to remain insensitive to the size of the object to be annotated should be implemented.<br />
<br />
6. Compared to other models, the model was shown to not perform as well for larger objects (see table 3). This is likely due to the fact that vertex location determination is done in a highly compressed (28x28) representation compared to the input image(224x224). For larger objects, bounding boxes are larger. Each vertex represents many pixels. When up-converted back to the input image/bounding box size these may lead to errors especially when considering a very precise evaluation metric (intersection over union) is used. Potentially, the results can be improved by considering a higher resolution for the internal representation or one that scales with the size of the bounding.<br />
<br />
=Code=<br />
# [https://github.com/AlexMa011/pytorch-polygon-rnn] (unofficial)<br />
# Code for an updated version of the model is available at [https://github.com/fidler-lab/polyrnn-pp] (official)</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=stat946w18/Towards_Image_Understanding_From_Deep_Compression_Without_Decoding&diff=40785stat946w18/Towards Image Understanding From Deep Compression Without Decoding2018-11-22T05:18:11Z<p>Gchalato: /* Learned Deeply Compressed Representations */</p>
<hr />
<div>Paper Title: Towards Image Understanding from Deep Compression Without Decoding - ICLR 2018<br />
<br />
Presented By: Aravind Ravi<br />
<br />
== Introduction ==<br />
Recent advances in the deep neural network (DNN) based image compression methods have shown potential improvements in image quality, savings in storage and bandwidth reduction. These methods leverage common neural network architectures such as convolutional autoencoders or recurrent neural networks to compress and reconstruct RGB images and outperform classical techniques such as JPEG2000 and BPG on perceptual metrics such as structural similarity index (SSIM) and multi-scale structural similarity index (MS-SSIM).<br />
<br />
These approaches encode an image <math>x </math> to some feature map (compressed representation), which is subsequently quantized to a set of symbols <math>z </math>. These symbols are then losslessly compressed to a bitstream, from which a decoder reconstructs an image <math>{\hat{x}} </math>, of the same dimensions as <math>x </math>.<br />
<br />
Learned compression algorithms have an advantage over engineering compression algorithms in that they can be much more easily adapted to specific domains. For example, a learned compression algorithm might be able to learn good performance on compressing medical images, without specifically tuning the algorithm.<br />
<br />
In this paper, the authors explore the idea of applying the learned representations to perform inference without reconstructing the compressed image. Specifically, instead of reconstructing an RGB image from the compressed representation and feeding it to a network for inference, the paper proposes to use a modified network that bypasses reconstruction of the RGB image.<br />
<br />
The rationale behind this approach is that the neural network architectures commonly used for learned compression (in particular the encoders) are similar to the ones commonly used for inference, and learned image encoders are hence, in principle, capable of extracting features relevant for inference tasks. The encoder might learn features relevant for inference purely by training on the compression task, and can be forced to learn these features by training on the compression and inference tasks jointly<br />
<br />
The advantage of learning an encoder for image compression which produces compressed representation containing features relevant for inference is obvious in scenarios where images are transmitted (e.g. from a mobile device) before processing (e.g. in the cloud), as it saves reconstruction of the RGB image as well as part of the feature extraction and hence speeds up processing. A typical use case is a cloud photo storage application where every image is processed immediately upon upload for indexing and search purposes.<br />
<br />
Note: [https://en.wikipedia.org/wiki/Structural_similarity More Information on SSIM, MSSIM]<br />
<br />
== Intuition ==<br />
<br />
Compression techniques (something as common as zipping) are commonly used by us in day to day file handling tasks. Most often we use engineered compression techniques. Deep Neural Networks (DNNs) are nonlinear function approximators which act as feature extractors, extracting features from inputs (like images or sound files). These can be seen as learning based compression techniques as they can perform compression and they can be trained using back propagation as well. If image classification can be done on these compressed files, large image data sets like hyperspectral images and MRI images can be stored efficiently and the compressed files can be used directly by the DNNs for classification or reinforcement learning tasks.<br />
<br />
==Motivation and Contributions==<br />
The authors propose to perform image understanding tasks such as image classification and segmentation directly on DNN based compressed representations. Performing the image understanding tasks on the compressed representations/encoded feature maps has two advantages. <br />
# This method bypasses the process of decoding the image into the RGB space before classification<br />
# The authors show that it reduces the overall computational complexity up to 2 times<br />
<br />
=== Contributions of the Paper ===<br />
* A method to perform image classification and semantic segmentation from compressed representations. In large scale image understanding problems, learning from a compressed representation is definitely something that is interesting. <br />
* The proposed method offers classification accuracy similar to that achieved on decompressed images while reducing the computational complexity by 2 times.<br />
* Semantic segmentation has been shown to be as accurate as performance on decompressed images for moderate compression rates and higher accuracy for aggressive compression rates. In addition, this method achieves lower computational complexity.<br />
* Joint training for image compression and classification has been shown to improve the quality of the image and increase in accuracy of classification and segmentation<br />
<br />
==Related Work==<br />
<br />
The prior work has shown image classification from compressed images based on engineered codecs. Some of the works in this area are:<br />
<br />
* In video analysis domain: Action recognition (Yeo et al., 2008; Kantorov & Laptev, 2014)<br />
* Classification of compressed hyperspectral images (Hahn et al., 2014; Aghagolzadeh & Radha, 2015)<br />
* Discrete Cosine Transform based compression performed on images before feeding into a neural network, which shows an improvement in training speed by up to 10 times Fu & Guimaraes (2016)<br />
* Video analysis on compressed video (using engineered codecs) has also been studied in the past (Babu et al., 2016)<br />
* Criticism on document image analysis methods (Javed et al.2017)<br />
<br />
The authors propose a method that does inference on top of learned feature representation and hence has a direct relation to unsupervised feature learning using autoencoders.<br />
They also claim that so far there hasn't been any work using learned compressed representations for image classification and segmentation.<br />
<br />
==Learned Deeply Compressed Representations==<br />
<br />
The image compression task is performed based on a convolutional autoencoder architecture proposed by Theis et al. 2017 (shown in the figure below), and a variant of the training procedure described by Agustsson et. al 2017. <br />
<br />
[[File:AR_theisAutoencoder.png|600px|center]]<br />
<br />
Some points to better understand the architecture:<br />
<br />
1. Most convolutions are done in a convolved, lower-dimensional space to speed up computation<br />
<br />
2. Different activation functions are used. Blank arrows indicate the identity function (no additional linearity), while black arrows indicate leaky rectifications<br />
<br />
3. The “round” box simply rounds all elements in the tensor to the nearest integer<br />
<br />
4. The “subpix” block is just an upsampling /reconstruction block where the feature map’s coefficients are reshuffled after a convolution<br />
<br />
<br />
<br />
=== Compression Architecture ===<br />
<br />
The compression network is an autoencoder that takes an input image <math>x </math> and outputs <math>{\hat{x}} </math> as the approximation to the input. <br />
<br />
[[File:AR_Fig2a.png|300px|center]]<br />
<br />
The encoder has the following structure: It starts with 2 convolutional layers with spatial subsampling by a factor of 2, followed by 3 residual units, and a final convolutional layer with spatial subsampling by a factor of 2. This results in a <math>w/8</math> x <math>h/8</math> x <math>C</math> dimensional representation, where <math>w </math> and <math>h </math> are the spatial dimensions of <math>x </math>, and the number of channels C is a hyperparameter related to the rate <math>R </math>. This representation is then quantized to a discrete set of symbols, forming a compressed representation, <math>z </math>.<br />
<br />
To get the reconstruction <math>{\hat{x}} </math>, the compressed representation is fed into the decoder, which mirrors the encoder, but uses upsampling and deconvolutions instead of subsampling and convolutions.<br />
<br />
Quantizing the compressed representation imposes a distortion <math>D </math> on <math>{\hat{x}} </math> w.r.t. <math>x </math>, i.e., it increases the reconstruction error. This is traded for a decrease in entropy of the quantized compressed representation<br />
<math>z </math> which leads to a decrease of the length of the bitstream as measured by the rate <math>R </math>. Thus, to train the image compression network, the classical rate-distortion trade-off <math>D + \beta R</math> is minimized. As a metric for <math>D </math>, the mean squared error (MSE) between <math>x </math> and <math>{\hat{x}} </math> are used and <math>R</math> is estimated using<br />
<math>H(q)</math>. <math>H(q)</math> is the entropy of the probability distribution over the symbols and is estimated using a histogram of the probability distribution (as done by Agustsson et al., 2017). The trade-off between MSE and the entropy is controlled by adjusting <math>\beta </math>. For each <math>\beta </math> an operating point is derived where the images have a certain bit rate, as measured by bits per pixel (bpp), and corresponding MSE. To better control the bpp, a target entropy Ht is introduced by the authors to formulate the loss defined as:<br />
<br />
\begin{align}<br />
\mathcal{L_c} = \text{MSE}(x,{\hat{x}})+\beta\max({H(q)}-{H_t},0)<br />
\end{align}<br />
<br />
Agustsson et. al 2017, proposed a method to overcome the issue of non-differentiability of the quantization step by proposing a differentiable approximation to the quantization. This method has been adapted to suit the current application in the paper.<br />
<br />
Three operating points at 0.0983 bpp (C=8), 0.330 bpp (C=16), and 0.635 bpp (C=32) are obtained empirically. All further experiments are performed with these three operating points and the results for the same are presented in the following sections.<br />
<br />
==Image Classification from Compressed Representations==<br />
<br />
=== Classification on RGB Images ===<br />
<br />
For the image classification task based on the RGB images, the authors use the ResNet-50 architecture. <br />
Further information on residual networks can be found in the following links: <br />
[https://youtu.be/K0uoBKBQ1gA ResNets Part-1]<br />
[https://youtu.be/GSsKdtoatm8 ResNets Part-2]<br />
<br />
The details of the architecture are presented in the table below:<br />
<br />
[[File:AR_Tab1.png|400px|center]]<br />
<br />
In this paper, the number of 14x14 (conv4_x) blocks have been modified to obtain a new architecture called ResNet-71. <br />
<br />
=== Classification on Compressed Representations ===<br />
<br />
For input images with spatial dimension 224x224, the encoder of the compression network outputs a compressed representation with dimensions 28x28xC, where C is the number of channels. To use this compressed representation as input to the classification network, a simple variant of the ResNet architecture is proposed. This variant is referred to as cResNet-k, where c stands for “compressed representation” and k is the<br />
number of convolutional layers in the network. These networks are constructed by simply “cutting off” the front of the regular (RGB) ResNet. The root-block of the network and the residual layers that have a larger spatial dimension than 28x28 are removed. To adjust the number of layers k, the ResNet architecture proposed by He et al. (2015) is used and the number of 14x14 (conv4 x) residual blocks are modified.<br />
<br />
In this way, three different architectures are derived:<br />
* cResNet-39 is ResNet-50 with the first 11 layers removed as described above, and this significantly reduces computational cost<br />
* cResNet-51<br />
* cResNet-72<br />
<br />
cResNet-51 and cResNet-72 are obtained by adding 14x14 residual blocks to match the computational cost of ResNet-50 and ResNet-71 respectively.<br />
<br />
The detailed description of all the network architectures are presented below:<br />
<br />
[[File:AR_Tab3.png|600px|center]]<br />
<br />
==Semantic Segmentation from Compressed Representations==<br />
<br />
For semantic segmentation, the ResNet based DeepLab architecture is adapted for the proposed application. The cResNet<br />
and ResNet image classification architectures are re-purposed with atrous<br />
convolutions, where the filters are upsampled instead of downsampling the feature maps. This is<br />
done to increase their receptive field and to prevent aggressive subsampling of the feature maps. For segmentation, the ResNet architecture is restructured such<br />
that the output feature map has 8 times smaller spatial dimension than the original RGB image (instead<br />
subsampling by a factor 32 times like for classification). When using the cResNets the output feature<br />
map has the same spatial dimensions as the input compressed representation (instead of subsampling<br />
4 times like for classification). This results in comparably sized feature maps for both the compressed<br />
representation and the reconstructed RGB images. Finally the last 1000-way classification layer of<br />
these classification architectures is replaced by an atrous spatial pyramid pooling (ASPP) with four<br />
parallel branches with rates {6, 12, 18, 24}, which provides the final pixel-wise classification.<br />
<br />
==Joint Training for Compression and Image Classification==<br />
<br />
The authors propose a joint training strategy to combine compression and classification tasks. To do this, the proposed method combines the compression network and the cResNet-51 architecture. The figure below shows the combined pipeline:<br />
<br />
[[File:AR_Fig2b.png|300px|center]]<br />
<br />
All parts, encoder, decoder, and inference network, are trained at the same time. The compressed representation is fed<br />
to the decoder to optimize for mean-squared reconstruction error and to a cResNet-51 network to<br />
optimize for classification using a cross-entropy loss. The combined loss function takes the form:<br />
<br />
\begin{align}<br />
\mathcal{L_c} = \gamma(\text{MSE}(x,{\hat{x}})+\beta\max({H(q)}-{H_t},0))+l_{ce}(y,{\hat{y}})<br />
\end{align}<br />
<br />
where the loss terms for the compression network, <math> \mathcal{L_c} = \text{MSE}(x,{\hat{x}})+\beta\max({H(q)}-{H_t},0)</math>, are the same as in training for compression only. <math> l_{ce}</math> is the cross-entropy loss for classification.<br />
<math>\gamma </math> controls the trade-off between the compression loss and the classification loss.<br />
<br />
==Experiments and Results==<br />
<br />
=== Learned Deeply Compressed Representations Results ===<br />
<br />
All experiments have been performed on the ILSVRC2012 dataset.<br />
<br />
The metrics used to measure the compression quality are as follows: <br />
* PSNR (Peak Signal-to-Noise Ratio) is a standard measure, depending monotonically on mean squared error defined as: <br />
<br />
\begin{align}<br />
PSNR = 10(\log_{10}(255^2/MSE))<br />
\end{align}<br />
<br />
* SSIM (Structural Similarity Index) and MS-SSIM (Multi-Scale SSIM) are metrics proposed to measure the similarity of images as perceived by humans<br />
<br />
The figure below depicts the performance of the deep compression models vs. standard JPEG and JPEG2000. Higher values are better. The proposed technique outperforms the JPEG and JPEC2000 at the operating points used in this paper.<br />
<br />
[[File:AR_Fig8.png|600px|center]]<br />
<br />
The learned compressed representations are illustrated in the figure below. <br />
<br />
[[File:AR_Fig9.png|500px|center]]<br />
<br />
In the above figure, the original RGB-image is shown along with compressed versions of the RGB image which are reconstructed from the compressed representations. The 4 channels with the highest entropy are shown in the visualizations. These visualizations indicate how the networks compress an image, as the rate (bpp) gets lower the entropy cost of the network forces the<br />
compressed representation to use fewer quantization levels, as can clearly be seen. For the most aggressive compression, the channel maps use only 2 levels for the compressed representation.<br />
<br />
=== Classification on Compressed Representations ===<br />
<br />
All experiments have been performed on the ILSVRC2012 dataset. It consists of 1.28 million training images and 50k validation images. These images are distributed across 1000 diverse classes. For image classification, the top-1 classification accuracy and top-5 classification accuracy are reported on the validation set on 224x224 center crops for RGB images and 28x28 center crops for the compressed representation.<br />
<br />
==== Training Procedure ====<br />
<br />
The compression network is fixed while training the classification network, both when training with compressed representations and with reconstructed compressed RGB images. For the compressed representations, the output of the fixed encoder (the compressed representation) is provided input to the cResNets (decoder is not needed). When training on the reconstructed compressed RGB images, the output of the fixed encoder-decoder (RGB image) is provided as input to the ResNet. This is done for each operating point.<br />
<br />
Refer to Appendix A Section A4, of the paper for details on the hyperparameters and optimization used for training the network [1].<br />
<br />
==== Classification Results ====<br />
<br />
The tables below present the results of the classification at each operating point, both classifying from the compressed representation and the corresponding reconstructed compressed RGB images.<br />
<br />
[[File:AR_Tab2.png|400|center]]<br />
<br />
Figure below shows the validation curves for ResNet-50, cResNet-51, and cResNet-39. <br />
<br />
[[File:AR_Fig3.png|700|center]]<br />
<br />
For the 2 classification architectures with the same computational complexity (ResNet-50 and cResNet-51), the validation curves at the 0.635 bpp compression operating point almost coincide, with ResNet-50 performing slightly better. As the rate (bpp) gets smaller this performance gap gets smaller. The table above shows the<br />
classification results when the different architectures have converged. At the 0.635 bpp operating point, ResNet-50 only performs 0.5% better in top-5 accuracy than cResNet-51, while for the 0.0983 bpp operating point this difference is only 0.3%.<br />
Using the same pre-processing and the same learning rate schedule but starting from the original uncompressed RGB images yields 89.96% top-5 accuracy. The top-5 accuracy obtained from the compressed representation at the 0.635 bpp compression operating point, 87.85%, is even competitive<br />
with that obtained for the original images at a significantly lower storage cost. Specifically, at 0.635 bpp the ImageNet dataset requires 24.8 GB of storage space instead of 144 GB for the original version, a reduction by a factor 5.8 times.<br />
<br />
Notes on top-1 and top-5 accuracy:<br />
<br />
* Top-1 accuracy: This is the conventional accuracy metric used in machine learning. Wherein if the true label of the input to a model matches the highest probability class of the last layer of the output of CNN (predicted class probability), then the given input is correctly classified, else it is considered as incorrectly classified.<br />
* Top-5 accuracy: In this case, if any of the model's 5 highest classification probabilities match with the true label of the input, then this is considered as a correct classification, else it is an incorrect classification.<br />
<br />
===Semantic Segmentation Results===<br />
<br />
All experiments have been performed on the PASCAL VOC-2012 dataset for semantic segmentation. It has 20 object foreground classes and 1 background class. The dataset<br />
consists of 1464 training and 1449 validation images. In every image, each pixel is annotated with<br />
one of the 20 + 1 classes. The original dataset is furthermore augmented with extra annotations, so the final dataset has 10,582 images for training and 1449 images for validation.<br />
<br />
All performance is measured on pixel wise intersection-over-union (IoU) averaged over all the classes or mean-intersection-over-union (mIoU) on the validation set. <br />
<br />
[https://www.pyimagesearch.com/2016/11/07/intersection-over-union-iou-for-object-detection/ Details on IoU]<br />
<br />
==== Training Procedure ====<br />
The cResNet/ResNet networks are pre-trained on the ImageNet dataset using the procedure described earlier on the image classification task, the encoder and decoder is fixed as in the earlier scenario. The architectures are then adapted with dilated convolutions, cResNet-d/ResNet-d, and<br />
finetuned on the semantic segmentation task.<br />
<br />
Refer to Appendix A Section A5, of the paper for details on the hyperparameters and optimization used for training the network [1].<br />
<br />
==== Segmentation Results ====<br />
<br />
The table below shows the mIoU results for the segmentation task.<br />
<br />
[[File:AR_Tab2.png|450|center]]<br />
<br />
The figure below illustrates the segmentation results with respect to each compression operating point.<br />
<br />
[[File:AR_Fig4.png|700|center]]<br />
<br />
For semantic segmentation ResNet-50-d and cResNet-51-d perform equally well at the 0.635 bpp compression operating point. For the<br />
0.330 bpp operating point, segmentation from the compressed representation performs slightly better, 0.37%, and at the 0.0983 bpp operating point segmentation from the compressed representation<br />
performs considerably better than for the reconstructed compressed RGB images, by 1.65%.<br />
<br />
[[File:AR_Fig5.png|600px|center]]<br />
<br />
The above figure shows the predicted segmentation visually for both the cResNet-51-d and the ResNet-50-d<br />
architecture at each operating point. Along with the segmentation, it also shows the original uncompressed<br />
RGB image and the reconstructed compressed RGB image. These images highlight<br />
the challenging nature of these segmentation tasks, but they can nevertheless be performed using the<br />
compressed representation. They also clearly indicate that the compression affects the segmentation,<br />
as lowering the rate (bpp) progressively removes details in the image. Comparing the segmentation<br />
from the reconstructed RGB images to the segmentation from the compressed representation visually,<br />
the performance is similar.<br />
<br />
The figure below is another example of visual results of segmentation from compressed representation and reconstructed RGB<br />
images. The performance is visually similar for all operating points except for the 0.0983<br />
bpp operating point where the reconstructed RGB image fails to capture the back part of<br />
the train, while the compressed representation manages to capture that aspect of the image in the<br />
segmentation.<br />
<br />
[[File:AR_Fig10.png|600px|center]]<br />
<br />
=== Results on Computational Gains ===<br />
<br />
[[File:AR_Fig6.png|400px|center]]<br />
<br />
=====Computational Gains on Classification=====<br />
<br />
The figure on the left illustrates, the top-5 classification accuracy as a function of computational<br />
complexity for the 0.0983 bpp compression operating point.<br />
Looking at a fixed computational cost, the reconstructed compressed RGB images perform about 0.25% better. Looking at a fixed classification cost, inference from the compressed representation costs about 0.6 * 10^9 FLOPs more. However when accounting for the decoding cost at a fixed<br />
classification performance, inference from the reconstructed compressed RGB images costs 2.2*10^9 FLOPs more than inference from the compressed representation.<br />
<br />
=====Computational Gains on Segmentation=====<br />
<br />
In the figure on the right illustrates, the mIoU validation performance is shown as a function of computational complexity for<br />
the 0.0983 bpp compression operating point. <br />
Here, even without accounting for the decoding cost of the reconstructed images, the compressed representation<br />
performs better. At a fixed computational cost, segmentation from the compressed representation gives about 0.7% better mIoU. And at a fixed mIoU the computational cost is about 3.3*10^9 FLOPs<br />
lower for compressed representations. Accounting for the decoding costs this difference becomes 6.1*10^9 FLOPs. due to the nature of the dilated convolutions and the increased feature map size, the<br />
relative computational gains for segmentation are not as pronounced as for classification.<br />
<br />
===Joint Training for Compression and Image Classification===<br />
<br />
==== Training Procedure ====<br />
<br />
When doing joint training, the compression network and the classification networks are first initialized<br />
from a trained state obtained as described previously. After initialization, the networks are<br />
both finetuned jointly. For a detailed<br />
description of hyperparameters used and the training schedule see Appendix A8.<br />
<br />
To control that the change in classification accuracy is not only due to (1) a better compression<br />
operating point or (2) the fact that the cResNet is trained longer, the following is done. A new operating point is obtained by finetuning the compression network only using the schedule described<br />
above. The cResNet-51 is trained on top of this new operating point from scratch. Finally, the compression network is fixed at the new operating point, and the cResNet-51 is trained for 9 epochs. <br />
<br />
To obtain segmentation results, the jointly trained network is used. The operating point is fixed and the jointly finetuned classification network is adopted fro segmentation (cResNet-51-d).<br />
<br />
==== Joint Training Results ====<br />
<br />
[[File:AR_Fig7.png|400px|center]]<br />
<br />
It can be seen from the figure, that the classification and segmentation results “move<br />
up” from the baseline through fine tuning. When training jointly the improvement for classification are larger and<br />
a significant improvement for segmentation is achieved. For the 0.635 bpp operating point the classification performance is similar for training the network jointly and training<br />
the compression network only, but when using these operating points for segmentation the difference is considerable.<br />
<br />
The results presented by the authors suggest an improvement in classification by 2%, a performance gain which would<br />
require an additional 75% of the computational complexity of cResNet-51. The segmentation<br />
performance after training the networks jointly is 1.7% better in mIoU than training only<br />
the compression network.<br />
<br />
==Critique==<br />
<br />
The paper proposes how previous work in auto-encoders and image compression can be extended effectively to a novel task of a combined image compression and recognition task. The work has provided extensive experimental evaluation and evidence that suggests that learned compressed representations can be effective in classification and segmentation tasks. While maintaining the performance of the techniques to state of the art performance, the authors show that the proposed method can offer significant computational gains. The applications of this can be in<br />
multimedia communication, wireless transmission of images, video surveillance on the mobile edge, etc. With the advent of 5G and other new wireless technologies, this method offers capabilities that can be utilized to conserve wireless bandwidth, savings on storage while retaining the perceptual quality of images.<br />
The joint training of compression and classification network provides some added advantages and also shows that at aggressive compression rates the performance in classification and segmentation can be improved significantly.<br />
<br />
The authors mention that the complexity of the current approach is still high in comparison with methods like JPEG or JPEG2000. They also mention that this can be overcome when the networks are trained and run on GPU's. Although this has been seen as a drawback, with subsequent improvements in physical hardware and more specialized deep learning platforms, the limitation of the current approach can be overcome. While the authors did thorough experiments and gave extensive results on compressed representations and their advantages, the idea itself is not very novel.Finally, in the light of providing extensive experimental contributions,<br />
the authors have written a quite lengthy paper. There are parts of the paper where the ideas have been repeated frequently, and this could've been avoided leading to a more well-balanced length of the article.<br />
<br />
==Conclusion==<br />
<br />
The paper proposes an inference task using compressed image representations without the need to decode for classification and semantic segmentation. The paper has successfully demonstrated through a set of rigorous experiments the approach<br />
for performing the intended tasks. The results show significant improvements in computational complexity while maintaining state of the art classification and segmentation performance. The authors also intend to explore other computer vision tasks based on using compressed representation as part of the future work. They also suggest that this could potentially lead to gaining a better understanding of the features/compressed representations learned by image compression networks leading to applications in unsupervised or semi-supervised learning.<br />
<br />
==References==<br />
# Torfason, R., Mentzer, F., Agustsson, E., Tschannen, M., Timofte, R., & Van Gool, L. (2018). Towards image understanding from deep compression without decoding. arXiv preprint arXiv:1803.06131.<br />
# Theis, L., Shi, W., Cunningham, A., & Huszár, F. (2017). Lossy image compression with compressive autoencoders. arXiv preprint arXiv:1703.00395.<br />
# Agustsson, E., Mentzer, F., Tschannen, M., Cavigelli, L., Timofte, R., Benini, L., & Gool, L. V. (2017). Soft-to-hard vector quantization for end-to-end learning compressible representations. In Advances in Neural Information Processing Systems (pp. 1141-1151).<br />
# He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 770-778).<br />
# Chen, L. C., Papandreou, G., Kokkinos, I., Murphy, K., & Yuille, A. L. (2018). Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs. IEEE transactions on pattern analysis and machine intelligence, 40(4), 834-848.</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=stat946w18/Towards_Image_Understanding_From_Deep_Compression_Without_Decoding&diff=40784stat946w18/Towards Image Understanding From Deep Compression Without Decoding2018-11-22T05:17:56Z<p>Gchalato: /* Learned Deeply Compressed Representations */</p>
<hr />
<div>Paper Title: Towards Image Understanding from Deep Compression Without Decoding - ICLR 2018<br />
<br />
Presented By: Aravind Ravi<br />
<br />
== Introduction ==<br />
Recent advances in the deep neural network (DNN) based image compression methods have shown potential improvements in image quality, savings in storage and bandwidth reduction. These methods leverage common neural network architectures such as convolutional autoencoders or recurrent neural networks to compress and reconstruct RGB images and outperform classical techniques such as JPEG2000 and BPG on perceptual metrics such as structural similarity index (SSIM) and multi-scale structural similarity index (MS-SSIM).<br />
<br />
These approaches encode an image <math>x </math> to some feature map (compressed representation), which is subsequently quantized to a set of symbols <math>z </math>. These symbols are then losslessly compressed to a bitstream, from which a decoder reconstructs an image <math>{\hat{x}} </math>, of the same dimensions as <math>x </math>.<br />
<br />
Learned compression algorithms have an advantage over engineering compression algorithms in that they can be much more easily adapted to specific domains. For example, a learned compression algorithm might be able to learn good performance on compressing medical images, without specifically tuning the algorithm.<br />
<br />
In this paper, the authors explore the idea of applying the learned representations to perform inference without reconstructing the compressed image. Specifically, instead of reconstructing an RGB image from the compressed representation and feeding it to a network for inference, the paper proposes to use a modified network that bypasses reconstruction of the RGB image.<br />
<br />
The rationale behind this approach is that the neural network architectures commonly used for learned compression (in particular the encoders) are similar to the ones commonly used for inference, and learned image encoders are hence, in principle, capable of extracting features relevant for inference tasks. The encoder might learn features relevant for inference purely by training on the compression task, and can be forced to learn these features by training on the compression and inference tasks jointly<br />
<br />
The advantage of learning an encoder for image compression which produces compressed representation containing features relevant for inference is obvious in scenarios where images are transmitted (e.g. from a mobile device) before processing (e.g. in the cloud), as it saves reconstruction of the RGB image as well as part of the feature extraction and hence speeds up processing. A typical use case is a cloud photo storage application where every image is processed immediately upon upload for indexing and search purposes.<br />
<br />
Note: [https://en.wikipedia.org/wiki/Structural_similarity More Information on SSIM, MSSIM]<br />
<br />
== Intuition ==<br />
<br />
Compression techniques (something as common as zipping) are commonly used by us in day to day file handling tasks. Most often we use engineered compression techniques. Deep Neural Networks (DNNs) are nonlinear function approximators which act as feature extractors, extracting features from inputs (like images or sound files). These can be seen as learning based compression techniques as they can perform compression and they can be trained using back propagation as well. If image classification can be done on these compressed files, large image data sets like hyperspectral images and MRI images can be stored efficiently and the compressed files can be used directly by the DNNs for classification or reinforcement learning tasks.<br />
<br />
==Motivation and Contributions==<br />
The authors propose to perform image understanding tasks such as image classification and segmentation directly on DNN based compressed representations. Performing the image understanding tasks on the compressed representations/encoded feature maps has two advantages. <br />
# This method bypasses the process of decoding the image into the RGB space before classification<br />
# The authors show that it reduces the overall computational complexity up to 2 times<br />
<br />
=== Contributions of the Paper ===<br />
* A method to perform image classification and semantic segmentation from compressed representations. In large scale image understanding problems, learning from a compressed representation is definitely something that is interesting. <br />
* The proposed method offers classification accuracy similar to that achieved on decompressed images while reducing the computational complexity by 2 times.<br />
* Semantic segmentation has been shown to be as accurate as performance on decompressed images for moderate compression rates and higher accuracy for aggressive compression rates. In addition, this method achieves lower computational complexity.<br />
* Joint training for image compression and classification has been shown to improve the quality of the image and increase in accuracy of classification and segmentation<br />
<br />
==Related Work==<br />
<br />
The prior work has shown image classification from compressed images based on engineered codecs. Some of the works in this area are:<br />
<br />
* In video analysis domain: Action recognition (Yeo et al., 2008; Kantorov & Laptev, 2014)<br />
* Classification of compressed hyperspectral images (Hahn et al., 2014; Aghagolzadeh & Radha, 2015)<br />
* Discrete Cosine Transform based compression performed on images before feeding into a neural network, which shows an improvement in training speed by up to 10 times Fu & Guimaraes (2016)<br />
* Video analysis on compressed video (using engineered codecs) has also been studied in the past (Babu et al., 2016)<br />
* Criticism on document image analysis methods (Javed et al.2017)<br />
<br />
The authors propose a method that does inference on top of learned feature representation and hence has a direct relation to unsupervised feature learning using autoencoders.<br />
They also claim that so far there hasn't been any work using learned compressed representations for image classification and segmentation.<br />
<br />
==Learned Deeply Compressed Representations==<br />
<br />
The image compression task is performed based on a convolutional autoencoder architecture proposed by Theis et al. 2017 (shown in the figure below), and a variant of the training procedure described by Agustsson et. al 2017. <br />
<br />
[[File:AR_theisAutoencoder.png|600px|center]]<br />
<br />
Some points to better understand the architecture:<br />
1. Most convolutions are done in a convolved, lower-dimensional space to speed up computation<br />
2. Different activation functions are used. Blank arrows indicate the identity function (no additional linearity), while black arrows indicate leaky rectifications<br />
3. The “round” box simply rounds all elements in the tensor to the nearest integer<br />
4. The “subpix” block is just an upsampling /reconstruction block where the feature map’s coefficients are reshuffled after a convolution<br />
<br />
<br />
<br />
=== Compression Architecture ===<br />
<br />
The compression network is an autoencoder that takes an input image <math>x </math> and outputs <math>{\hat{x}} </math> as the approximation to the input. <br />
<br />
[[File:AR_Fig2a.png|300px|center]]<br />
<br />
The encoder has the following structure: It starts with 2 convolutional layers with spatial subsampling by a factor of 2, followed by 3 residual units, and a final convolutional layer with spatial subsampling by a factor of 2. This results in a <math>w/8</math> x <math>h/8</math> x <math>C</math> dimensional representation, where <math>w </math> and <math>h </math> are the spatial dimensions of <math>x </math>, and the number of channels C is a hyperparameter related to the rate <math>R </math>. This representation is then quantized to a discrete set of symbols, forming a compressed representation, <math>z </math>.<br />
<br />
To get the reconstruction <math>{\hat{x}} </math>, the compressed representation is fed into the decoder, which mirrors the encoder, but uses upsampling and deconvolutions instead of subsampling and convolutions.<br />
<br />
Quantizing the compressed representation imposes a distortion <math>D </math> on <math>{\hat{x}} </math> w.r.t. <math>x </math>, i.e., it increases the reconstruction error. This is traded for a decrease in entropy of the quantized compressed representation<br />
<math>z </math> which leads to a decrease of the length of the bitstream as measured by the rate <math>R </math>. Thus, to train the image compression network, the classical rate-distortion trade-off <math>D + \beta R</math> is minimized. As a metric for <math>D </math>, the mean squared error (MSE) between <math>x </math> and <math>{\hat{x}} </math> are used and <math>R</math> is estimated using<br />
<math>H(q)</math>. <math>H(q)</math> is the entropy of the probability distribution over the symbols and is estimated using a histogram of the probability distribution (as done by Agustsson et al., 2017). The trade-off between MSE and the entropy is controlled by adjusting <math>\beta </math>. For each <math>\beta </math> an operating point is derived where the images have a certain bit rate, as measured by bits per pixel (bpp), and corresponding MSE. To better control the bpp, a target entropy Ht is introduced by the authors to formulate the loss defined as:<br />
<br />
\begin{align}<br />
\mathcal{L_c} = \text{MSE}(x,{\hat{x}})+\beta\max({H(q)}-{H_t},0)<br />
\end{align}<br />
<br />
Agustsson et. al 2017, proposed a method to overcome the issue of non-differentiability of the quantization step by proposing a differentiable approximation to the quantization. This method has been adapted to suit the current application in the paper.<br />
<br />
Three operating points at 0.0983 bpp (C=8), 0.330 bpp (C=16), and 0.635 bpp (C=32) are obtained empirically. All further experiments are performed with these three operating points and the results for the same are presented in the following sections.<br />
<br />
==Image Classification from Compressed Representations==<br />
<br />
=== Classification on RGB Images ===<br />
<br />
For the image classification task based on the RGB images, the authors use the ResNet-50 architecture. <br />
Further information on residual networks can be found in the following links: <br />
[https://youtu.be/K0uoBKBQ1gA ResNets Part-1]<br />
[https://youtu.be/GSsKdtoatm8 ResNets Part-2]<br />
<br />
The details of the architecture are presented in the table below:<br />
<br />
[[File:AR_Tab1.png|400px|center]]<br />
<br />
In this paper, the number of 14x14 (conv4_x) blocks have been modified to obtain a new architecture called ResNet-71. <br />
<br />
=== Classification on Compressed Representations ===<br />
<br />
For input images with spatial dimension 224x224, the encoder of the compression network outputs a compressed representation with dimensions 28x28xC, where C is the number of channels. To use this compressed representation as input to the classification network, a simple variant of the ResNet architecture is proposed. This variant is referred to as cResNet-k, where c stands for “compressed representation” and k is the<br />
number of convolutional layers in the network. These networks are constructed by simply “cutting off” the front of the regular (RGB) ResNet. The root-block of the network and the residual layers that have a larger spatial dimension than 28x28 are removed. To adjust the number of layers k, the ResNet architecture proposed by He et al. (2015) is used and the number of 14x14 (conv4 x) residual blocks are modified.<br />
<br />
In this way, three different architectures are derived:<br />
* cResNet-39 is ResNet-50 with the first 11 layers removed as described above, and this significantly reduces computational cost<br />
* cResNet-51<br />
* cResNet-72<br />
<br />
cResNet-51 and cResNet-72 are obtained by adding 14x14 residual blocks to match the computational cost of ResNet-50 and ResNet-71 respectively.<br />
<br />
The detailed description of all the network architectures are presented below:<br />
<br />
[[File:AR_Tab3.png|600px|center]]<br />
<br />
==Semantic Segmentation from Compressed Representations==<br />
<br />
For semantic segmentation, the ResNet based DeepLab architecture is adapted for the proposed application. The cResNet<br />
and ResNet image classification architectures are re-purposed with atrous<br />
convolutions, where the filters are upsampled instead of downsampling the feature maps. This is<br />
done to increase their receptive field and to prevent aggressive subsampling of the feature maps. For segmentation, the ResNet architecture is restructured such<br />
that the output feature map has 8 times smaller spatial dimension than the original RGB image (instead<br />
subsampling by a factor 32 times like for classification). When using the cResNets the output feature<br />
map has the same spatial dimensions as the input compressed representation (instead of subsampling<br />
4 times like for classification). This results in comparably sized feature maps for both the compressed<br />
representation and the reconstructed RGB images. Finally the last 1000-way classification layer of<br />
these classification architectures is replaced by an atrous spatial pyramid pooling (ASPP) with four<br />
parallel branches with rates {6, 12, 18, 24}, which provides the final pixel-wise classification.<br />
<br />
==Joint Training for Compression and Image Classification==<br />
<br />
The authors propose a joint training strategy to combine compression and classification tasks. To do this, the proposed method combines the compression network and the cResNet-51 architecture. The figure below shows the combined pipeline:<br />
<br />
[[File:AR_Fig2b.png|300px|center]]<br />
<br />
All parts, encoder, decoder, and inference network, are trained at the same time. The compressed representation is fed<br />
to the decoder to optimize for mean-squared reconstruction error and to a cResNet-51 network to<br />
optimize for classification using a cross-entropy loss. The combined loss function takes the form:<br />
<br />
\begin{align}<br />
\mathcal{L_c} = \gamma(\text{MSE}(x,{\hat{x}})+\beta\max({H(q)}-{H_t},0))+l_{ce}(y,{\hat{y}})<br />
\end{align}<br />
<br />
where the loss terms for the compression network, <math> \mathcal{L_c} = \text{MSE}(x,{\hat{x}})+\beta\max({H(q)}-{H_t},0)</math>, are the same as in training for compression only. <math> l_{ce}</math> is the cross-entropy loss for classification.<br />
<math>\gamma </math> controls the trade-off between the compression loss and the classification loss.<br />
<br />
==Experiments and Results==<br />
<br />
=== Learned Deeply Compressed Representations Results ===<br />
<br />
All experiments have been performed on the ILSVRC2012 dataset.<br />
<br />
The metrics used to measure the compression quality are as follows: <br />
* PSNR (Peak Signal-to-Noise Ratio) is a standard measure, depending monotonically on mean squared error defined as: <br />
<br />
\begin{align}<br />
PSNR = 10(\log_{10}(255^2/MSE))<br />
\end{align}<br />
<br />
* SSIM (Structural Similarity Index) and MS-SSIM (Multi-Scale SSIM) are metrics proposed to measure the similarity of images as perceived by humans<br />
<br />
The figure below depicts the performance of the deep compression models vs. standard JPEG and JPEG2000. Higher values are better. The proposed technique outperforms the JPEG and JPEC2000 at the operating points used in this paper.<br />
<br />
[[File:AR_Fig8.png|600px|center]]<br />
<br />
The learned compressed representations are illustrated in the figure below. <br />
<br />
[[File:AR_Fig9.png|500px|center]]<br />
<br />
In the above figure, the original RGB-image is shown along with compressed versions of the RGB image which are reconstructed from the compressed representations. The 4 channels with the highest entropy are shown in the visualizations. These visualizations indicate how the networks compress an image, as the rate (bpp) gets lower the entropy cost of the network forces the<br />
compressed representation to use fewer quantization levels, as can clearly be seen. For the most aggressive compression, the channel maps use only 2 levels for the compressed representation.<br />
<br />
=== Classification on Compressed Representations ===<br />
<br />
All experiments have been performed on the ILSVRC2012 dataset. It consists of 1.28 million training images and 50k validation images. These images are distributed across 1000 diverse classes. For image classification, the top-1 classification accuracy and top-5 classification accuracy are reported on the validation set on 224x224 center crops for RGB images and 28x28 center crops for the compressed representation.<br />
<br />
==== Training Procedure ====<br />
<br />
The compression network is fixed while training the classification network, both when training with compressed representations and with reconstructed compressed RGB images. For the compressed representations, the output of the fixed encoder (the compressed representation) is provided input to the cResNets (decoder is not needed). When training on the reconstructed compressed RGB images, the output of the fixed encoder-decoder (RGB image) is provided as input to the ResNet. This is done for each operating point.<br />
<br />
Refer to Appendix A Section A4, of the paper for details on the hyperparameters and optimization used for training the network [1].<br />
<br />
==== Classification Results ====<br />
<br />
The tables below present the results of the classification at each operating point, both classifying from the compressed representation and the corresponding reconstructed compressed RGB images.<br />
<br />
[[File:AR_Tab2.png|400|center]]<br />
<br />
Figure below shows the validation curves for ResNet-50, cResNet-51, and cResNet-39. <br />
<br />
[[File:AR_Fig3.png|700|center]]<br />
<br />
For the 2 classification architectures with the same computational complexity (ResNet-50 and cResNet-51), the validation curves at the 0.635 bpp compression operating point almost coincide, with ResNet-50 performing slightly better. As the rate (bpp) gets smaller this performance gap gets smaller. The table above shows the<br />
classification results when the different architectures have converged. At the 0.635 bpp operating point, ResNet-50 only performs 0.5% better in top-5 accuracy than cResNet-51, while for the 0.0983 bpp operating point this difference is only 0.3%.<br />
Using the same pre-processing and the same learning rate schedule but starting from the original uncompressed RGB images yields 89.96% top-5 accuracy. The top-5 accuracy obtained from the compressed representation at the 0.635 bpp compression operating point, 87.85%, is even competitive<br />
with that obtained for the original images at a significantly lower storage cost. Specifically, at 0.635 bpp the ImageNet dataset requires 24.8 GB of storage space instead of 144 GB for the original version, a reduction by a factor 5.8 times.<br />
<br />
Notes on top-1 and top-5 accuracy:<br />
<br />
* Top-1 accuracy: This is the conventional accuracy metric used in machine learning. Wherein if the true label of the input to a model matches the highest probability class of the last layer of the output of CNN (predicted class probability), then the given input is correctly classified, else it is considered as incorrectly classified.<br />
* Top-5 accuracy: In this case, if any of the model's 5 highest classification probabilities match with the true label of the input, then this is considered as a correct classification, else it is an incorrect classification.<br />
<br />
===Semantic Segmentation Results===<br />
<br />
All experiments have been performed on the PASCAL VOC-2012 dataset for semantic segmentation. It has 20 object foreground classes and 1 background class. The dataset<br />
consists of 1464 training and 1449 validation images. In every image, each pixel is annotated with<br />
one of the 20 + 1 classes. The original dataset is furthermore augmented with extra annotations, so the final dataset has 10,582 images for training and 1449 images for validation.<br />
<br />
All performance is measured on pixel wise intersection-over-union (IoU) averaged over all the classes or mean-intersection-over-union (mIoU) on the validation set. <br />
<br />
[https://www.pyimagesearch.com/2016/11/07/intersection-over-union-iou-for-object-detection/ Details on IoU]<br />
<br />
==== Training Procedure ====<br />
The cResNet/ResNet networks are pre-trained on the ImageNet dataset using the procedure described earlier on the image classification task, the encoder and decoder is fixed as in the earlier scenario. The architectures are then adapted with dilated convolutions, cResNet-d/ResNet-d, and<br />
finetuned on the semantic segmentation task.<br />
<br />
Refer to Appendix A Section A5, of the paper for details on the hyperparameters and optimization used for training the network [1].<br />
<br />
==== Segmentation Results ====<br />
<br />
The table below shows the mIoU results for the segmentation task.<br />
<br />
[[File:AR_Tab2.png|450|center]]<br />
<br />
The figure below illustrates the segmentation results with respect to each compression operating point.<br />
<br />
[[File:AR_Fig4.png|700|center]]<br />
<br />
For semantic segmentation ResNet-50-d and cResNet-51-d perform equally well at the 0.635 bpp compression operating point. For the<br />
0.330 bpp operating point, segmentation from the compressed representation performs slightly better, 0.37%, and at the 0.0983 bpp operating point segmentation from the compressed representation<br />
performs considerably better than for the reconstructed compressed RGB images, by 1.65%.<br />
<br />
[[File:AR_Fig5.png|600px|center]]<br />
<br />
The above figure shows the predicted segmentation visually for both the cResNet-51-d and the ResNet-50-d<br />
architecture at each operating point. Along with the segmentation, it also shows the original uncompressed<br />
RGB image and the reconstructed compressed RGB image. These images highlight<br />
the challenging nature of these segmentation tasks, but they can nevertheless be performed using the<br />
compressed representation. They also clearly indicate that the compression affects the segmentation,<br />
as lowering the rate (bpp) progressively removes details in the image. Comparing the segmentation<br />
from the reconstructed RGB images to the segmentation from the compressed representation visually,<br />
the performance is similar.<br />
<br />
The figure below is another example of visual results of segmentation from compressed representation and reconstructed RGB<br />
images. The performance is visually similar for all operating points except for the 0.0983<br />
bpp operating point where the reconstructed RGB image fails to capture the back part of<br />
the train, while the compressed representation manages to capture that aspect of the image in the<br />
segmentation.<br />
<br />
[[File:AR_Fig10.png|600px|center]]<br />
<br />
=== Results on Computational Gains ===<br />
<br />
[[File:AR_Fig6.png|400px|center]]<br />
<br />
=====Computational Gains on Classification=====<br />
<br />
The figure on the left illustrates, the top-5 classification accuracy as a function of computational<br />
complexity for the 0.0983 bpp compression operating point.<br />
Looking at a fixed computational cost, the reconstructed compressed RGB images perform about 0.25% better. Looking at a fixed classification cost, inference from the compressed representation costs about 0.6 * 10^9 FLOPs more. However when accounting for the decoding cost at a fixed<br />
classification performance, inference from the reconstructed compressed RGB images costs 2.2*10^9 FLOPs more than inference from the compressed representation.<br />
<br />
=====Computational Gains on Segmentation=====<br />
<br />
In the figure on the right illustrates, the mIoU validation performance is shown as a function of computational complexity for<br />
the 0.0983 bpp compression operating point. <br />
Here, even without accounting for the decoding cost of the reconstructed images, the compressed representation<br />
performs better. At a fixed computational cost, segmentation from the compressed representation gives about 0.7% better mIoU. And at a fixed mIoU the computational cost is about 3.3*10^9 FLOPs<br />
lower for compressed representations. Accounting for the decoding costs this difference becomes 6.1*10^9 FLOPs. due to the nature of the dilated convolutions and the increased feature map size, the<br />
relative computational gains for segmentation are not as pronounced as for classification.<br />
<br />
===Joint Training for Compression and Image Classification===<br />
<br />
==== Training Procedure ====<br />
<br />
When doing joint training, the compression network and the classification networks are first initialized<br />
from a trained state obtained as described previously. After initialization, the networks are<br />
both finetuned jointly. For a detailed<br />
description of hyperparameters used and the training schedule see Appendix A8.<br />
<br />
To control that the change in classification accuracy is not only due to (1) a better compression<br />
operating point or (2) the fact that the cResNet is trained longer, the following is done. A new operating point is obtained by finetuning the compression network only using the schedule described<br />
above. The cResNet-51 is trained on top of this new operating point from scratch. Finally, the compression network is fixed at the new operating point, and the cResNet-51 is trained for 9 epochs. <br />
<br />
To obtain segmentation results, the jointly trained network is used. The operating point is fixed and the jointly finetuned classification network is adopted fro segmentation (cResNet-51-d).<br />
<br />
==== Joint Training Results ====<br />
<br />
[[File:AR_Fig7.png|400px|center]]<br />
<br />
It can be seen from the figure, that the classification and segmentation results “move<br />
up” from the baseline through fine tuning. When training jointly the improvement for classification are larger and<br />
a significant improvement for segmentation is achieved. For the 0.635 bpp operating point the classification performance is similar for training the network jointly and training<br />
the compression network only, but when using these operating points for segmentation the difference is considerable.<br />
<br />
The results presented by the authors suggest an improvement in classification by 2%, a performance gain which would<br />
require an additional 75% of the computational complexity of cResNet-51. The segmentation<br />
performance after training the networks jointly is 1.7% better in mIoU than training only<br />
the compression network.<br />
<br />
==Critique==<br />
<br />
The paper proposes how previous work in auto-encoders and image compression can be extended effectively to a novel task of a combined image compression and recognition task. The work has provided extensive experimental evaluation and evidence that suggests that learned compressed representations can be effective in classification and segmentation tasks. While maintaining the performance of the techniques to state of the art performance, the authors show that the proposed method can offer significant computational gains. The applications of this can be in<br />
multimedia communication, wireless transmission of images, video surveillance on the mobile edge, etc. With the advent of 5G and other new wireless technologies, this method offers capabilities that can be utilized to conserve wireless bandwidth, savings on storage while retaining the perceptual quality of images.<br />
The joint training of compression and classification network provides some added advantages and also shows that at aggressive compression rates the performance in classification and segmentation can be improved significantly.<br />
<br />
The authors mention that the complexity of the current approach is still high in comparison with methods like JPEG or JPEG2000. They also mention that this can be overcome when the networks are trained and run on GPU's. Although this has been seen as a drawback, with subsequent improvements in physical hardware and more specialized deep learning platforms, the limitation of the current approach can be overcome. While the authors did thorough experiments and gave extensive results on compressed representations and their advantages, the idea itself is not very novel.Finally, in the light of providing extensive experimental contributions,<br />
the authors have written a quite lengthy paper. There are parts of the paper where the ideas have been repeated frequently, and this could've been avoided leading to a more well-balanced length of the article.<br />
<br />
==Conclusion==<br />
<br />
The paper proposes an inference task using compressed image representations without the need to decode for classification and semantic segmentation. The paper has successfully demonstrated through a set of rigorous experiments the approach<br />
for performing the intended tasks. The results show significant improvements in computational complexity while maintaining state of the art classification and segmentation performance. The authors also intend to explore other computer vision tasks based on using compressed representation as part of the future work. They also suggest that this could potentially lead to gaining a better understanding of the features/compressed representations learned by image compression networks leading to applications in unsupervised or semi-supervised learning.<br />
<br />
==References==<br />
# Torfason, R., Mentzer, F., Agustsson, E., Tschannen, M., Timofte, R., & Van Gool, L. (2018). Towards image understanding from deep compression without decoding. arXiv preprint arXiv:1803.06131.<br />
# Theis, L., Shi, W., Cunningham, A., & Huszár, F. (2017). Lossy image compression with compressive autoencoders. arXiv preprint arXiv:1703.00395.<br />
# Agustsson, E., Mentzer, F., Tschannen, M., Cavigelli, L., Timofte, R., Benini, L., & Gool, L. V. (2017). Soft-to-hard vector quantization for end-to-end learning compressible representations. In Advances in Neural Information Processing Systems (pp. 1141-1151).<br />
# He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 770-778).<br />
# Chen, L. C., Papandreou, G., Kokkinos, I., Murphy, K., & Yuille, A. L. (2018). Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs. IEEE transactions on pattern analysis and machine intelligence, 40(4), 834-848.</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Co-Teaching&diff=40751Co-Teaching2018-11-22T02:53:23Z<p>Gchalato: /* Bootstrap */</p>
<hr />
<div>=Introduction=<br />
==Title of Paper==<br />
Co-teaching: Robust Training Deep Neural Networks with Extremely Noisy Labels<br />
==Contributions==<br />
The paper proposes a novel approach to training deep neural networks on data with noisy labels. The proposed architecture, named ‘co-teaching’, maintains two networks simultaneously, which focuses on training on selected clean instances and avoids estimating the noise transition matrix. In addition, using stochastic optimization with momentum to train the deep networks and clean data can be memorized by nonlinear deep networks, which becomes robust. The experiments are conducted on noisy versions of MNIST, CIFAR-10 and CIFAR-100 datasets. Empirical results demonstrate that, under extremely noisy circumstances (i.e., 45% of noisy labels), the robustness of deep learning models trained by the Co-teaching approach is much superior to state-of-the-art baselines<br />
<br />
==Terminology==<br />
Ground-Truth Labels: The proper objective labels (i.e. the real, or ‘true’, labels) of the data. <br />
<br />
Noisy Labels: Labels that are corrupted (either manually or through the data collection process) from ground-truth labels. This can result in false positives.<br />
<br />
=Intuition=<br />
The Co-teaching architecture maintains two networks with different learning abilities simultaneously. The reason why Co-teaching is more robust can be explained as follows. Usually when learning on a batch of noisy data, only the error from the network itself is transferred back to facilitate learning. But in the case of Co-teaching, the two networks are able to filter different type of errors, and flow back to itself and the other network. As a result, the two models learn together, from the network itself and the partner network.<br />
<br />
=Motivation=<br />
The paper draws motivation from two key facts:<br />
• That many data collection processes yield noisy labels. <br />
• That deep neural networks have a high capacity to overfit to noisy labels. <br />
Because of these facts, it is challenging to train deep networks to be robust with noisy labels. <br />
=Related Works=<br />
<br />
1. Statistical learning methods: Some approaches use statistical learning methods for the problem of learning from extremely noisy labels. These approaches can be divided into 3 strands: surrogate loss, noise estimation, and probabilistic modeling. In the surrogate loss category, one work proposes an unbiased estimator to provide the noise corrected loss approach. Another work presented a robust non-convex loss, which is the special case in a family of robust losses. In the noise rate estimation category, some authors propose a class-probability estimator using order statistics on the range of scores. Another work presented the same estimator using the slope of ROC curve. In the probabilistic modeling category, there is a two coin model proposed to handle noise labels from multiple annotators. <br />
<br />
2. Deep learning methods: There are also deep learning approaches that can be used to approach data with noisy labels. One work proposed a unified framework to distill knowledge from clean labels and knowledge graphs. Another work trained a label cleaning network by a small set of clean labels and used it to reduce the noise in large-scale noisy labels. There is also a proposed joint optimization framework to learn parameters and estimate true labels simultaneously. <br />
Another work leverages an additional validation set to adaptively assign weights to training examples in every iteration. One particular paper ads a crowd layer after the output layer for noisy labels from multiple annotators. <br />
<br />
3. Learning to teach methods: It is another approach to this problem. The methods are made up by the teacher and student networks. The teacher network selects more informative instances for better training of student networks. Most works did not account for noisy labels, with exception to MentorNet, which applied the idea on data with noisy labels.<br />
<br />
=Co-Teaching Algorithm=<br />
<br />
[[File:Co-Teaching_Algorithm.png|600px|center]]<br />
<br />
The idea as shown in the algorithm above is to train two deep networks simultaneously. In each mini-batch using mini-batch gradient descent, each network selects its small-loss instances as useful knowledge and then teaches these useful instances to the peer network. <math>R(T)</math> governs the percentage of small-loss instances to be used in updating the parameters of each network.<br />
<br />
=Summary of Experiment=<br />
==Proposed Method==<br />
The proposed co-teaching method maintains two networks simultaneously, and samples instances with small loss at each mini batch. The sample of small-loss instances is then taught to the peer network. <br />
[[File:Co-Teaching Fig 1.png|600px|center]] <br />
The co-teaching method relies on research that suggests deep networks learn clean and easy patterns in initial epochs, but are susceptible to overfitting noisy labels as the number of epochs grows. To counteract this, the co-teaching method reduces the mini-batch size by gradually increasing a drop rate (i.e., noisy instances with higher loss will be dropped at an increasing rate). <br />
The mini-batches are swapped between peer networks due to the underlying intuition that different classifiers will generate different decision boundaries. Swapping the mini-batches constitutes a sort of ‘peer-reviewing’ that promotes noise reduction since the error from a network is not directly transferred back to itself. <br />
==Dataset Corruption==<br />
To simulate learning with noisy labels, the datasets (which are clean by default) are manually corrupted by applying a noise transformation matrix. Two methods are used for generating such noise transformation matrices: pair flipping and symmetry. <br />
[[File:Co-Teaching Fig 2.png|600px|center]] <br />
Three noise conditions are simulated for comparing co-teaching with baseline methods.<br />
<br />
Note: Corruption of Dataset here means randomly choosing a wrong label instead of the target label by applying noise. <br />
<br />
{| class="wikitable"<br />
{| border="1" cellpadding="3"<br />
|-<br />
|width="60pt"|Method<br />
|width="100pt"|Noise Rate<br />
|width="700pt"|Rationale<br />
|-<br />
| Pair Flipping || 45% || Almost half of the instances have noisy labels. Simulates erroneous labels which are similar to true labels. <br />
|-<br />
| Symmetry || 50% || Half of the instances have noisy labels. Labels have a constant probability of being corrupted. Further rationale can be found at [1].<br />
|-<br />
| Symmetry || 20% || Verify the robustness of co-teaching in a low-level noise scenario. <br />
|}<br />
|}<br />
<br />
==Baseline Comparisons==<br />
The co-teaching method is compared with several baseline approaches, which have varying:<br />
• proficiency in dealing with a large number of classes,<br />
• ability to resist heavy noise,<br />
• need to combine with specific network architectures, and<br />
• need to be pretrained. <br />
<br />
[[File:Co-Teaching Fig 3.png|600px|center]] <br />
===Bootstrap===<br />
The general idea behind bootstrapping is to dynamically change (correct) noisy labels during training. The idea is to take a value derived from the original and predicted class. The final label is some convex combination of the two. It should be noted that the weighting of the prediction is increased over time to account for the model itself improving. Of course, this procedure needs to be finely tuned to prevent it from rampantly changing correct labels before it becomes accurate. [2].<br />
<br />
===S-Model===<br />
Using an additional softmax layer to model the noise transition matrix [3].<br />
===F-Correction===<br />
Correcting the prediction by using a noise transition matrix which is estimated by a standard network [4].<br />
===Decoupling===<br />
Two separate classifiers are used in this technique. Parameters are updated using only the samples that are classified differently between the two models [5].<br />
===MentorNet===<br />
A mentor network is weights the probability of data instances being clean/noisy in order to train the student network on cleaner instances [6].<br />
<br />
==Implementation Details==<br />
Two CNN models using the same architecture (shown below) are used as the peer networks for the co-teaching method. They are initialized with different parameters in order to be significantly different from one another (different initial parameters can lead to different local minima). An Adam optimizer (momentum=0.9), a learning rate of 0.001, a batch size of 128, and 200 epochs are used for each dataset. The networks also utilize dropout and batch normalization. <br />
<br />
[[File: Co-Teaching Table 3.png|center]] <br />
=Results and Discussion=<br />
The co-teaching algorithm is compared to the baseline approaches under the noise conditions previously described. The results are as follows. <br />
==MNIST==<br />
The results of testing on the MNIST dataset are shown below. The Symmetry-20% case can be taken as a near-baseline; all methods perform well. However, under the Symmetry-50% case, all methods except MentorNet and Co-Teaching drop below 90% accuracy. Under the Pair-45% case, all methods except MentorNet and Co-Teaching drop below 60%. Under both high-noise conditions, the Co-Teaching method produces the highest accuracy. Similar patterns can be seen in the two additional sets of test results, though the specific accuracy values are different. Co-Teaching performs best under the high-noise situations<br />
<br />
The images labelled 'Figure 3' show test accuracy with respect to epoch of the various algorithms. Many algorithms show evidence of over-fitting or being influenced by noisy data, after reaching initial high accuracy. MentorNet and Co-Teaching experience this less than other methods, and Co-Teaching generally achieves higher accuracy than MentorNet. <br />
<br />
Robustness of the proposed method to noise which plays an important rule in the evaluation, is evident in the plots which is better or comparable to the other methods.<br />
<br />
[[File:Co-Teaching Table 4.png|550px|center]]<br />
<br />
[[File:Co-Teaching Graphs MNIST.PNG|center]]<br />
<br />
==CIFAR10==<br />
[[File:Co-Teaching Table 5.png|550px|center]] <br />
<br />
[[File:Co-Teaching Graphs CIFAR10.PNG|center]]<br />
==CIFAR100==<br />
[[File:Co-Teaching Table 6.png|550px|center]] <br />
<br />
[[File: Co-Teaching Graphs CIFAR100.PNG|center]]<br />
==Choice of R(T) and <math> \tau</math>==<br />
There were some principles they followed when it came to choosing R(T) and <math> \tau</math>. R(T)=1, there was no instance needed at the beginning. They could safely update parameters in the early stage using the whole noise data since the deep neural networks would not memorize the noisy data. However, they need to drop more instances at the later stage. Because the model would eventually try to fit noisy data.<br />
<br />
R(T)=1-<math> \tau </math> *min{<math>T^{c}/T_{k},1 </math>} with <math> \tau=\epsilon </math>, where <math> \epsilon </math> is noise level.<br />
In this case, we consider c={0.5,1,2}. From Table 7, the test accuracy is stable.<br />
<br />
[[File: Co-Teaching Table 7.png|550px|center]] <br />
<br />
For <math> \tau</math>, we consider <math> \tau={0.5,0.75,1,1.25,1.5}\epsilon</math>. From Table 8, the performance can be improved with dropping more instances.<br />
[[File: Co-Teaching Table 8.png|550px|center]]<br />
<br />
=Conclusions=<br />
The main goal of the paper is to introduce the “Co-teaching” learning paradigm that uses two deep neural networks learning simultaneously to avoid noisy labels. Experiments are performed on several datasets such as MNIST, CIFAR-10, and CIFAR-100. The performance varied depending on the noise level in different scenarios. In the simulated ‘extreme noise’ scenarios, (pair-45% and symmetry-50%), the co-teaching methods outperforms baseline methods in terms of accuracy. This suggests that the co-teaching method is superior to the baseline methods in scenarios of extreme noise. The co-teaching method also performs competitively in the low-noise scenario (symmetry-20%).<br />
<br />
=Critique=<br />
==Lack of Task Diversity==<br />
The datasets used in this experiment are all image classification tasks – these results may not generalize to other deep learning applications, such as classifications from data with lower or higher dimensionality. <br />
==Needs to be expanded to other weak supervisions (Mentioned in conclusion)==<br />
Adaptation of the co-teaching method to train under other weak supervision (e.g. positive and unlabeled data) could expand the applicability of the paradigm. <br />
==Lack of Theoretical Development (Mentioned in conclusion)==<br />
This paper lacks any theoretical guarantees for co-teaching. Proving that the results shown in this study are generalizable would bolster the findings significantly. <br />
=References=<br />
[1] B. Van Rooyen, A. Menon, and B. Williamson. Learning with symmetric label noise: The<br />
importance of being unhinged. In NIPS, 2015.<br />
<br />
[2] S. Reed, H. Lee, D. Anguelov, C. Szegedy, D. Erhan, and A. Rabinovich. Training deep neural<br />
networks on noisy labels with bootstrapping. In ICLR, 2015.<br />
<br />
[3] J. Goldberger and E. Ben-Reuven. Training deep neural-networks using a noise adaptation layer.<br />
In ICLR, 2017.<br />
<br />
[4] G. Patrini, A. Rozza, A. Menon, R. Nock, and L. Qu. Making deep neural networks robust to<br />
label noise: A loss correction approach. In CVPR, 2017.<br />
<br />
[5] E. Malach and S. Shalev-Shwartz. Decoupling" when to update" from" how to update". In<br />
NIPS, 2017.<br />
<br />
[6] L. Jiang, Z. Zhou, T. Leung, L. Li, and L. Fei-Fei. Mentornet: Learning data-driven curriculum<br />
for very deep neural networks on corrupted labels. In ICML, 2018.</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Co-Teaching&diff=40750Co-Teaching2018-11-22T02:52:37Z<p>Gchalato: /* Bootstrap */</p>
<hr />
<div>=Introduction=<br />
==Title of Paper==<br />
Co-teaching: Robust Training Deep Neural Networks with Extremely Noisy Labels<br />
==Contributions==<br />
The paper proposes a novel approach to training deep neural networks on data with noisy labels. The proposed architecture, named ‘co-teaching’, maintains two networks simultaneously, which focuses on training on selected clean instances and avoids estimating the noise transition matrix. In addition, using stochastic optimization with momentum to train the deep networks and clean data can be memorized by nonlinear deep networks, which becomes robust. The experiments are conducted on noisy versions of MNIST, CIFAR-10 and CIFAR-100 datasets. Empirical results demonstrate that, under extremely noisy circumstances (i.e., 45% of noisy labels), the robustness of deep learning models trained by the Co-teaching approach is much superior to state-of-the-art baselines<br />
<br />
==Terminology==<br />
Ground-Truth Labels: The proper objective labels (i.e. the real, or ‘true’, labels) of the data. <br />
<br />
Noisy Labels: Labels that are corrupted (either manually or through the data collection process) from ground-truth labels. This can result in false positives.<br />
<br />
=Intuition=<br />
The Co-teaching architecture maintains two networks with different learning abilities simultaneously. The reason why Co-teaching is more robust can be explained as follows. Usually when learning on a batch of noisy data, only the error from the network itself is transferred back to facilitate learning. But in the case of Co-teaching, the two networks are able to filter different type of errors, and flow back to itself and the other network. As a result, the two models learn together, from the network itself and the partner network.<br />
<br />
=Motivation=<br />
The paper draws motivation from two key facts:<br />
• That many data collection processes yield noisy labels. <br />
• That deep neural networks have a high capacity to overfit to noisy labels. <br />
Because of these facts, it is challenging to train deep networks to be robust with noisy labels. <br />
=Related Works=<br />
<br />
1. Statistical learning methods: Some approaches use statistical learning methods for the problem of learning from extremely noisy labels. These approaches can be divided into 3 strands: surrogate loss, noise estimation, and probabilistic modeling. In the surrogate loss category, one work proposes an unbiased estimator to provide the noise corrected loss approach. Another work presented a robust non-convex loss, which is the special case in a family of robust losses. In the noise rate estimation category, some authors propose a class-probability estimator using order statistics on the range of scores. Another work presented the same estimator using the slope of ROC curve. In the probabilistic modeling category, there is a two coin model proposed to handle noise labels from multiple annotators. <br />
<br />
2. Deep learning methods: There are also deep learning approaches that can be used to approach data with noisy labels. One work proposed a unified framework to distill knowledge from clean labels and knowledge graphs. Another work trained a label cleaning network by a small set of clean labels and used it to reduce the noise in large-scale noisy labels. There is also a proposed joint optimization framework to learn parameters and estimate true labels simultaneously. <br />
Another work leverages an additional validation set to adaptively assign weights to training examples in every iteration. One particular paper ads a crowd layer after the output layer for noisy labels from multiple annotators. <br />
<br />
3. Learning to teach methods: It is another approach to this problem. The methods are made up by the teacher and student networks. The teacher network selects more informative instances for better training of student networks. Most works did not account for noisy labels, with exception to MentorNet, which applied the idea on data with noisy labels.<br />
<br />
=Co-Teaching Algorithm=<br />
<br />
[[File:Co-Teaching_Algorithm.png|600px|center]]<br />
<br />
The idea as shown in the algorithm above is to train two deep networks simultaneously. In each mini-batch using mini-batch gradient descent, each network selects its small-loss instances as useful knowledge and then teaches these useful instances to the peer network. <math>R(T)</math> governs the percentage of small-loss instances to be used in updating the parameters of each network.<br />
<br />
=Summary of Experiment=<br />
==Proposed Method==<br />
The proposed co-teaching method maintains two networks simultaneously, and samples instances with small loss at each mini batch. The sample of small-loss instances is then taught to the peer network. <br />
[[File:Co-Teaching Fig 1.png|600px|center]] <br />
The co-teaching method relies on research that suggests deep networks learn clean and easy patterns in initial epochs, but are susceptible to overfitting noisy labels as the number of epochs grows. To counteract this, the co-teaching method reduces the mini-batch size by gradually increasing a drop rate (i.e., noisy instances with higher loss will be dropped at an increasing rate). <br />
The mini-batches are swapped between peer networks due to the underlying intuition that different classifiers will generate different decision boundaries. Swapping the mini-batches constitutes a sort of ‘peer-reviewing’ that promotes noise reduction since the error from a network is not directly transferred back to itself. <br />
==Dataset Corruption==<br />
To simulate learning with noisy labels, the datasets (which are clean by default) are manually corrupted by applying a noise transformation matrix. Two methods are used for generating such noise transformation matrices: pair flipping and symmetry. <br />
[[File:Co-Teaching Fig 2.png|600px|center]] <br />
Three noise conditions are simulated for comparing co-teaching with baseline methods.<br />
<br />
Note: Corruption of Dataset here means randomly choosing a wrong label instead of the target label by applying noise. <br />
<br />
{| class="wikitable"<br />
{| border="1" cellpadding="3"<br />
|-<br />
|width="60pt"|Method<br />
|width="100pt"|Noise Rate<br />
|width="700pt"|Rationale<br />
|-<br />
| Pair Flipping || 45% || Almost half of the instances have noisy labels. Simulates erroneous labels which are similar to true labels. <br />
|-<br />
| Symmetry || 50% || Half of the instances have noisy labels. Labels have a constant probability of being corrupted. Further rationale can be found at [1].<br />
|-<br />
| Symmetry || 20% || Verify the robustness of co-teaching in a low-level noise scenario. <br />
|}<br />
|}<br />
<br />
==Baseline Comparisons==<br />
The co-teaching method is compared with several baseline approaches, which have varying:<br />
• proficiency in dealing with a large number of classes,<br />
• ability to resist heavy noise,<br />
• need to combine with specific network architectures, and<br />
• need to be pretrained. <br />
<br />
[[File:Co-Teaching Fig 3.png|600px|center]] <br />
===Bootstrap===<br />
The general idea behind bootstrapping is to dynamically change (correct) noisy labels during training. The idea is to take a weighted value derived from the original and predicted class. The final label is then some combination of the two. It should be noted that the weighting of the prediction is increased over time to account for the model itself improving. Of course, this procedure needs to be finely tuned to prevent it from rampantly changing correct labels before it becomes accurate. [2].<br />
<br />
===S-Model===<br />
Using an additional softmax layer to model the noise transition matrix [3].<br />
===F-Correction===<br />
Correcting the prediction by using a noise transition matrix which is estimated by a standard network [4].<br />
===Decoupling===<br />
Two separate classifiers are used in this technique. Parameters are updated using only the samples that are classified differently between the two models [5].<br />
===MentorNet===<br />
A mentor network is weights the probability of data instances being clean/noisy in order to train the student network on cleaner instances [6].<br />
<br />
==Implementation Details==<br />
Two CNN models using the same architecture (shown below) are used as the peer networks for the co-teaching method. They are initialized with different parameters in order to be significantly different from one another (different initial parameters can lead to different local minima). An Adam optimizer (momentum=0.9), a learning rate of 0.001, a batch size of 128, and 200 epochs are used for each dataset. The networks also utilize dropout and batch normalization. <br />
<br />
[[File: Co-Teaching Table 3.png|center]] <br />
=Results and Discussion=<br />
The co-teaching algorithm is compared to the baseline approaches under the noise conditions previously described. The results are as follows. <br />
==MNIST==<br />
The results of testing on the MNIST dataset are shown below. The Symmetry-20% case can be taken as a near-baseline; all methods perform well. However, under the Symmetry-50% case, all methods except MentorNet and Co-Teaching drop below 90% accuracy. Under the Pair-45% case, all methods except MentorNet and Co-Teaching drop below 60%. Under both high-noise conditions, the Co-Teaching method produces the highest accuracy. Similar patterns can be seen in the two additional sets of test results, though the specific accuracy values are different. Co-Teaching performs best under the high-noise situations<br />
<br />
The images labelled 'Figure 3' show test accuracy with respect to epoch of the various algorithms. Many algorithms show evidence of over-fitting or being influenced by noisy data, after reaching initial high accuracy. MentorNet and Co-Teaching experience this less than other methods, and Co-Teaching generally achieves higher accuracy than MentorNet. <br />
<br />
Robustness of the proposed method to noise which plays an important rule in the evaluation, is evident in the plots which is better or comparable to the other methods.<br />
<br />
[[File:Co-Teaching Table 4.png|550px|center]]<br />
<br />
[[File:Co-Teaching Graphs MNIST.PNG|center]]<br />
<br />
==CIFAR10==<br />
[[File:Co-Teaching Table 5.png|550px|center]] <br />
<br />
[[File:Co-Teaching Graphs CIFAR10.PNG|center]]<br />
==CIFAR100==<br />
[[File:Co-Teaching Table 6.png|550px|center]] <br />
<br />
[[File: Co-Teaching Graphs CIFAR100.PNG|center]]<br />
==Choice of R(T) and <math> \tau</math>==<br />
There were some principles they followed when it came to choosing R(T) and <math> \tau</math>. R(T)=1, there was no instance needed at the beginning. They could safely update parameters in the early stage using the whole noise data since the deep neural networks would not memorize the noisy data. However, they need to drop more instances at the later stage. Because the model would eventually try to fit noisy data.<br />
<br />
R(T)=1-<math> \tau </math> *min{<math>T^{c}/T_{k},1 </math>} with <math> \tau=\epsilon </math>, where <math> \epsilon </math> is noise level.<br />
In this case, we consider c={0.5,1,2}. From Table 7, the test accuracy is stable.<br />
<br />
[[File: Co-Teaching Table 7.png|550px|center]] <br />
<br />
For <math> \tau</math>, we consider <math> \tau={0.5,0.75,1,1.25,1.5}\epsilon</math>. From Table 8, the performance can be improved with dropping more instances.<br />
[[File: Co-Teaching Table 8.png|550px|center]]<br />
<br />
=Conclusions=<br />
The main goal of the paper is to introduce the “Co-teaching” learning paradigm that uses two deep neural networks learning simultaneously to avoid noisy labels. Experiments are performed on several datasets such as MNIST, CIFAR-10, and CIFAR-100. The performance varied depending on the noise level in different scenarios. In the simulated ‘extreme noise’ scenarios, (pair-45% and symmetry-50%), the co-teaching methods outperforms baseline methods in terms of accuracy. This suggests that the co-teaching method is superior to the baseline methods in scenarios of extreme noise. The co-teaching method also performs competitively in the low-noise scenario (symmetry-20%).<br />
<br />
=Critique=<br />
==Lack of Task Diversity==<br />
The datasets used in this experiment are all image classification tasks – these results may not generalize to other deep learning applications, such as classifications from data with lower or higher dimensionality. <br />
==Needs to be expanded to other weak supervisions (Mentioned in conclusion)==<br />
Adaptation of the co-teaching method to train under other weak supervision (e.g. positive and unlabeled data) could expand the applicability of the paradigm. <br />
==Lack of Theoretical Development (Mentioned in conclusion)==<br />
This paper lacks any theoretical guarantees for co-teaching. Proving that the results shown in this study are generalizable would bolster the findings significantly. <br />
=References=<br />
[1] B. Van Rooyen, A. Menon, and B. Williamson. Learning with symmetric label noise: The<br />
importance of being unhinged. In NIPS, 2015.<br />
<br />
[2] S. Reed, H. Lee, D. Anguelov, C. Szegedy, D. Erhan, and A. Rabinovich. Training deep neural<br />
networks on noisy labels with bootstrapping. In ICLR, 2015.<br />
<br />
[3] J. Goldberger and E. Ben-Reuven. Training deep neural-networks using a noise adaptation layer.<br />
In ICLR, 2017.<br />
<br />
[4] G. Patrini, A. Rozza, A. Menon, R. Nock, and L. Qu. Making deep neural networks robust to<br />
label noise: A loss correction approach. In CVPR, 2017.<br />
<br />
[5] E. Malach and S. Shalev-Shwartz. Decoupling" when to update" from" how to update". In<br />
NIPS, 2017.<br />
<br />
[6] L. Jiang, Z. Zhou, T. Leung, L. Li, and L. Fei-Fei. Mentornet: Learning data-driven curriculum<br />
for very deep neural networks on corrupted labels. In ICML, 2018.</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Co-Teaching&diff=40749Co-Teaching2018-11-22T02:31:58Z<p>Gchalato: /* Contributions */</p>
<hr />
<div>=Introduction=<br />
==Title of Paper==<br />
Co-teaching: Robust Training Deep Neural Networks with Extremely Noisy Labels<br />
==Contributions==<br />
The paper proposes a novel approach to training deep neural networks on data with noisy labels. The proposed architecture, named ‘co-teaching’, maintains two networks simultaneously, which focuses on training on selected clean instances and avoids estimating the noise transition matrix. In addition, using stochastic optimization with momentum to train the deep networks and clean data can be memorized by nonlinear deep networks, which becomes robust. The experiments are conducted on noisy versions of MNIST, CIFAR-10 and CIFAR-100 datasets. Empirical results demonstrate that, under extremely noisy circumstances (i.e., 45% of noisy labels), the robustness of deep learning models trained by the Co-teaching approach is much superior to state-of-the-art baselines<br />
<br />
==Terminology==<br />
Ground-Truth Labels: The proper objective labels (i.e. the real, or ‘true’, labels) of the data. <br />
<br />
Noisy Labels: Labels that are corrupted (either manually or through the data collection process) from ground-truth labels. This can result in false positives.<br />
<br />
=Intuition=<br />
The Co-teaching architecture maintains two networks with different learning abilities simultaneously. The reason why Co-teaching is more robust can be explained as follows. Usually when learning on a batch of noisy data, only the error from the network itself is transferred back to facilitate learning. But in the case of Co-teaching, the two networks are able to filter different type of errors, and flow back to itself and the other network. As a result, the two models learn together, from the network itself and the partner network.<br />
<br />
=Motivation=<br />
The paper draws motivation from two key facts:<br />
• That many data collection processes yield noisy labels. <br />
• That deep neural networks have a high capacity to overfit to noisy labels. <br />
Because of these facts, it is challenging to train deep networks to be robust with noisy labels. <br />
=Related Works=<br />
<br />
1. Statistical learning methods: Some approaches use statistical learning methods for the problem of learning from extremely noisy labels. These approaches can be divided into 3 strands: surrogate loss, noise estimation, and probabilistic modeling. In the surrogate loss category, one work proposes an unbiased estimator to provide the noise corrected loss approach. Another work presented a robust non-convex loss, which is the special case in a family of robust losses. In the noise rate estimation category, some authors propose a class-probability estimator using order statistics on the range of scores. Another work presented the same estimator using the slope of ROC curve. In the probabilistic modeling category, there is a two coin model proposed to handle noise labels from multiple annotators. <br />
<br />
2. Deep learning methods: There are also deep learning approaches that can be used to approach data with noisy labels. One work proposed a unified framework to distill knowledge from clean labels and knowledge graphs. Another work trained a label cleaning network by a small set of clean labels and used it to reduce the noise in large-scale noisy labels. There is also a proposed joint optimization framework to learn parameters and estimate true labels simultaneously. <br />
Another work leverages an additional validation set to adaptively assign weights to training examples in every iteration. One particular paper ads a crowd layer after the output layer for noisy labels from multiple annotators. <br />
<br />
3. Learning to teach methods: It is another approach to this problem. The methods are made up by the teacher and student networks. The teacher network selects more informative instances for better training of student networks. Most works did not account for noisy labels, with exception to MentorNet, which applied the idea on data with noisy labels.<br />
<br />
=Co-Teaching Algorithm=<br />
<br />
[[File:Co-Teaching_Algorithm.png|600px|center]]<br />
<br />
The idea as shown in the algorithm above is to train two deep networks simultaneously. In each mini-batch using mini-batch gradient descent, each network selects its small-loss instances as useful knowledge and then teaches these useful instances to the peer network. <math>R(T)</math> governs the percentage of small-loss instances to be used in updating the parameters of each network.<br />
<br />
=Summary of Experiment=<br />
==Proposed Method==<br />
The proposed co-teaching method maintains two networks simultaneously, and samples instances with small loss at each mini batch. The sample of small-loss instances is then taught to the peer network. <br />
[[File:Co-Teaching Fig 1.png|600px|center]] <br />
The co-teaching method relies on research that suggests deep networks learn clean and easy patterns in initial epochs, but are susceptible to overfitting noisy labels as the number of epochs grows. To counteract this, the co-teaching method reduces the mini-batch size by gradually increasing a drop rate (i.e., noisy instances with higher loss will be dropped at an increasing rate). <br />
The mini-batches are swapped between peer networks due to the underlying intuition that different classifiers will generate different decision boundaries. Swapping the mini-batches constitutes a sort of ‘peer-reviewing’ that promotes noise reduction since the error from a network is not directly transferred back to itself. <br />
==Dataset Corruption==<br />
To simulate learning with noisy labels, the datasets (which are clean by default) are manually corrupted by applying a noise transformation matrix. Two methods are used for generating such noise transformation matrices: pair flipping and symmetry. <br />
[[File:Co-Teaching Fig 2.png|600px|center]] <br />
Three noise conditions are simulated for comparing co-teaching with baseline methods.<br />
<br />
Note: Corruption of Dataset here means randomly choosing a wrong label instead of the target label by applying noise. <br />
<br />
{| class="wikitable"<br />
{| border="1" cellpadding="3"<br />
|-<br />
|width="60pt"|Method<br />
|width="100pt"|Noise Rate<br />
|width="700pt"|Rationale<br />
|-<br />
| Pair Flipping || 45% || Almost half of the instances have noisy labels. Simulates erroneous labels which are similar to true labels. <br />
|-<br />
| Symmetry || 50% || Half of the instances have noisy labels. Labels have a constant probability of being corrupted. Further rationale can be found at [1].<br />
|-<br />
| Symmetry || 20% || Verify the robustness of co-teaching in a low-level noise scenario. <br />
|}<br />
|}<br />
<br />
==Baseline Comparisons==<br />
The co-teaching method is compared with several baseline approaches, which have varying:<br />
• proficiency in dealing with a large number of classes,<br />
• ability to resist heavy noise,<br />
• need to combine with specific network architectures, and<br />
• need to be pretrained. <br />
<br />
[[File:Co-Teaching Fig 3.png|600px|center]] <br />
===Bootstrap===<br />
A method that deems a weighted combination of predicted and original labels as correct, and then solves kernels by backpropagation [2].<br />
===S-Model===<br />
Using an additional softmax layer to model the noise transition matrix [3].<br />
===F-Correction===<br />
Correcting the prediction by using a noise transition matrix which is estimated by a standard network [4].<br />
===Decoupling===<br />
Two separate classifiers are used in this technique. Parameters are updated using only the samples that are classified differently between the two models [5].<br />
===MentorNet===<br />
A mentor network is weights the probability of data instances being clean/noisy in order to train the student network on cleaner instances [6].<br />
<br />
==Implementation Details==<br />
Two CNN models using the same architecture (shown below) are used as the peer networks for the co-teaching method. They are initialized with different parameters in order to be significantly different from one another (different initial parameters can lead to different local minima). An Adam optimizer (momentum=0.9), a learning rate of 0.001, a batch size of 128, and 200 epochs are used for each dataset. The networks also utilize dropout and batch normalization. <br />
<br />
[[File: Co-Teaching Table 3.png|center]] <br />
=Results and Discussion=<br />
The co-teaching algorithm is compared to the baseline approaches under the noise conditions previously described. The results are as follows. <br />
==MNIST==<br />
The results of testing on the MNIST dataset are shown below. The Symmetry-20% case can be taken as a near-baseline; all methods perform well. However, under the Symmetry-50% case, all methods except MentorNet and Co-Teaching drop below 90% accuracy. Under the Pair-45% case, all methods except MentorNet and Co-Teaching drop below 60%. Under both high-noise conditions, the Co-Teaching method produces the highest accuracy. Similar patterns can be seen in the two additional sets of test results, though the specific accuracy values are different. Co-Teaching performs best under the high-noise situations<br />
<br />
The images labelled 'Figure 3' show test accuracy with respect to epoch of the various algorithms. Many algorithms show evidence of over-fitting or being influenced by noisy data, after reaching initial high accuracy. MentorNet and Co-Teaching experience this less than other methods, and Co-Teaching generally achieves higher accuracy than MentorNet. <br />
<br />
Robustness of the proposed method to noise which plays an important rule in the evaluation, is evident in the plots which is better or comparable to the other methods.<br />
<br />
[[File:Co-Teaching Table 4.png|550px|center]]<br />
<br />
[[File:Co-Teaching Graphs MNIST.PNG|center]]<br />
<br />
==CIFAR10==<br />
[[File:Co-Teaching Table 5.png|550px|center]] <br />
<br />
[[File:Co-Teaching Graphs CIFAR10.PNG|center]]<br />
==CIFAR100==<br />
[[File:Co-Teaching Table 6.png|550px|center]] <br />
<br />
[[File: Co-Teaching Graphs CIFAR100.PNG|center]]<br />
==Choice of R(T) and <math> \tau</math>==<br />
There were some principles they followed when it came to choosing R(T) and <math> \tau</math>. R(T)=1, there was no instance needed at the beginning. They could safely update parameters in the early stage using the whole noise data since the deep neural networks would not memorize the noisy data. However, they need to drop more instances at the later stage. Because the model would eventually try to fit noisy data.<br />
<br />
R(T)=1-<math> \tau </math> *min{<math>T^{c}/T_{k},1 </math>} with <math> \tau=\epsilon </math>, where <math> \epsilon </math> is noise level.<br />
In this case, we consider c={0.5,1,2}. From Table 7, the test accuracy is stable.<br />
<br />
[[File: Co-Teaching Table 7.png|550px|center]] <br />
<br />
For <math> \tau</math>, we consider <math> \tau={0.5,0.75,1,1.25,1.5}\epsilon</math>. From Table 8, the performance can be improved with dropping more instances.<br />
[[File: Co-Teaching Table 8.png|550px|center]]<br />
<br />
=Conclusions=<br />
The main goal of the paper is to introduce the “Co-teaching” learning paradigm that uses two deep neural networks learning simultaneously to avoid noisy labels. Experiments are performed on several datasets such as MNIST, CIFAR-10, and CIFAR-100. The performance varied depending on the noise level in different scenarios. In the simulated ‘extreme noise’ scenarios, (pair-45% and symmetry-50%), the co-teaching methods outperforms baseline methods in terms of accuracy. This suggests that the co-teaching method is superior to the baseline methods in scenarios of extreme noise. The co-teaching method also performs competitively in the low-noise scenario (symmetry-20%).<br />
<br />
=Critique=<br />
==Lack of Task Diversity==<br />
The datasets used in this experiment are all image classification tasks – these results may not generalize to other deep learning applications, such as classifications from data with lower or higher dimensionality. <br />
==Needs to be expanded to other weak supervisions (Mentioned in conclusion)==<br />
Adaptation of the co-teaching method to train under other weak supervision (e.g. positive and unlabeled data) could expand the applicability of the paradigm. <br />
==Lack of Theoretical Development (Mentioned in conclusion)==<br />
This paper lacks any theoretical guarantees for co-teaching. Proving that the results shown in this study are generalizable would bolster the findings significantly. <br />
=References=<br />
[1] B. Van Rooyen, A. Menon, and B. Williamson. Learning with symmetric label noise: The<br />
importance of being unhinged. In NIPS, 2015.<br />
<br />
[2] S. Reed, H. Lee, D. Anguelov, C. Szegedy, D. Erhan, and A. Rabinovich. Training deep neural<br />
networks on noisy labels with bootstrapping. In ICLR, 2015.<br />
<br />
[3] J. Goldberger and E. Ben-Reuven. Training deep neural-networks using a noise adaptation layer.<br />
In ICLR, 2017.<br />
<br />
[4] G. Patrini, A. Rozza, A. Menon, R. Nock, and L. Qu. Making deep neural networks robust to<br />
label noise: A loss correction approach. In CVPR, 2017.<br />
<br />
[5] E. Malach and S. Shalev-Shwartz. Decoupling" when to update" from" how to update". In<br />
NIPS, 2017.<br />
<br />
[6] L. Jiang, Z. Zhou, T. Leung, L. Li, and L. Fei-Fei. Mentornet: Learning data-driven curriculum<br />
for very deep neural networks on corrupted labels. In ICML, 2018.</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Learning_to_Teach&diff=40748Learning to Teach2018-11-22T02:28:02Z<p>Gchalato: /* Experiments */</p>
<hr />
<div><br />
<br />
=Introduction=<br />
<br />
This paper proposed the "learning to teach" (L2T) framework with two intelligent agents: a student model/agent, corresponding to the learner in traditional machine learning algorithms, and a teacher model/agent, determining the appropriate data, loss function, and hypothesis space to facilitate the learning of the student model.<br />
<br />
In modern human society, the role of teaching is heavily implicated in our education system; the goal is to equip students with the necessary knowledge and skills in an efficient manner. This is the fundamental ''student'' and ''teacher'' framework on which education stands. However, in the field of artificial intelligence (AI) and specifically machine learning, researchers have focused most of their efforts on the ''student'' (ie. designing various optimization algorithms to enhance the learning ability of intelligent agents). The paper argues that a formal study on the role of ‘teaching’ in AI is required. Analogous to teaching in human society, the teaching framework can: select training data that corresponds to the appropriate teaching materials (e.g. textbooks selected for the right difficulty), design loss functions that correspond to targeted examinations, and define the hypothesis space that corresponds to imparting the proper methodologies. Furthermore, an optimization framework (instead of heuristics) should be used to update the teaching skills based on the feedback from students, so as to achieve teacher-student co-evolution.<br />
<br />
Thus, the training phase of L2T would have several episodes of interactions between the teacher and the student model. Based on the state information in each step, the teacher model would update the teaching actions so that the student model could perform better on the Machine Learning problem. The student model would then provide reward signals back to the teacher model. These reward signals are used by the teacher model as part of the Reinforcement Learning process to update its parameters. This process is end-to-end trainable and the authors are convinced that once converged, the teacher model could be applied to new learning scenarios and even new students, without extra efforts on re-training.<br />
<br />
To demonstrate the practical value of the proposed approach, the '''training data scheduling''' problem is chosen as an example. The authors show that by using the proposed method to adaptively select the most<br />
suitable training data, they can significantly improve the accuracy and convergence speed of various neural networks including multi-layer perceptron (MLP), convolutional neural networks (CNNs)<br />
and recurrent neural networks (RNNs), for different applications including image classification and text understanding.<br />
<br />
=Related Work=<br />
The L2T framework connects with two emerging trends in machine learning. The first is the movement from simple to advanced learning. This includes meta-learning (Schmidhuber, 1987; Thrun & Pratt, 2012) which explores automatic learning by transferring learned knowledge from meta tasks [1]. This approach has been applied to few-shot learning scenarios and in designing general optimizers and neural network architectures. (Hochreiter et al., 2001; Andrychowicz et al., 2016; Li & Malik, 2016; Zoph & Le, 2017)<br />
<br />
The second is the teaching, which can be classified into either machine-teaching (Zhu, 2015) [2] or hardness based methods. The former seeks to construct a minimal training set for the student to learn a target model (ie. an oracle). The latter assumes an order of data from easy instances to hard ones, hardness being determined in different ways. In curriculum learning (CL) (Bengio et al, 2009; Spitkovsky et al. 2010; Tsvetkov et al, 2016) [3] measures hardness through heuristics of the data while self-paced learning (SPL) (Kumar et al., 2010; Lee & Grauman, 2011; Jiang et al., 2014; Supancic & Ramanan, 2013) [4] measures hardness by loss on data. <br />
<br />
The limitations of these works include the lack of a formally defined teaching problem, and the reliance on heuristics and fixed rules, which hinders generalization of the teaching task.<br />
<br />
=Learning to Teach=<br />
To introduce the problem and framework, without loss of generality, consider the setting of supervised learning.<br />
<br />
In supervised learning, each sample <math>x</math> is from a fixed but unknown distribution <math>P(x)</math>, and the corresponding label <math> y </math> is from a fixed but unknown distribution <math>P(y|x) </math>. The goal is to find a function <math>f_\omega(x)</math> with parameter vector <math>\omega</math> that minimizes the gap between the predicted label and the actual label.<br />
<br />
<br />
<br />
==Problem Definition==<br />
The student model, denoted &mu;(), takes the set of training data <math> D </math>, the function class <math> Ω </math>, and loss function <math> L </math> as input to output a function, <math> f(ω) </math>, with parameter <math>ω^*</math> which minimizes risk <math>R(ω)</math> as in:<br />
<br />
\begin{align*}<br />
ω^* = arg min_{w \in \Omega} \sum_{x,y \in D} L(y, f_ω(x)) =: \mu (D, L, \Omega)<br />
\end{align*}<br />
<br />
The teaching model, denoted φ, tries to provide <math> D </math>, <math> L </math>, and <math> Ω </math> (or any combination, denoted <math> A </math>) to the student model such that the student model either achieves lower risk R(ω) or progresses as fast as possible.<br />
<br />
::'''Training Data''': Outputting a good training set <math> D </math>, analogous to human teachers providing students with proper learning materials such as textbooks.<br />
::'''Loss Function''': Designing a good loss function <math> L </math> , analogous to providing useful assessment criteria for students.<br />
::'''Hypothesis Space''': Defining a good function class <math> Ω </math> which the student model can select from. This is analogous to human teachers providing appropriate context, eg. middle school students taught math with basic algebra while undergraduate students are taught with calculus. Different Ω leads to different errors and optimization problem (Mohri et al., 2012).<br />
<br />
==Framework==<br />
The training phase consists of the teacher providing the student with the subset <math> A_{train} </math> of <math> A </math> and then taking feedback to improve its own parameters. The L2T process is outlined in figure below:<br />
<br />
[[File: L2T_process.png | 500px|center]]<br />
<br />
* <math> s_t &isin; S </math> represents information available to the teacher model at time <math> t </math>. <math> s_t </math> is typically constructed from the current student model <math> f_{t−1} </math> and the past teaching history of the teacher model. <math> S </math> represents the set of states.<br />
* <math> a_t &isin; A </math> represents action taken the teacher model at time <math> t </math>, given state <math>s_t</math>. <math> A </math> represents the set of actions, where the action(s) can be any combination of teaching tasks involving the training data, loss function, and hypothesis space. <br />
* <math> φ_θ : S → A </math> is policy used by the teacher model to generate its action <math> φ_θ(s_t) = a_t </math><br />
* Student model takes <math> a_t </math> as input and outputs function <math> f_t </math>, by using the conventional ML techniques.<br />
<br />
Once the training process converges, the teacher model may be utilized to teach a different subset of <math> A </math> or teach a different student model.<br />
<br />
=Application=<br />
<br />
There are different approaches to training the teacher model, this paper will apply reinforcement learning with <math> φ_θ </math> being the ''policy'' that interacts with <math> S </math>, the ''environment''. The paper applies data teaching to train a deep neural network student, <math> f </math>, for several classification tasks. Thus the student feedback measure will be classification accuracy. Its learning rule will be mini-batch stochastic gradient descent, where batches of data will arrive sequentially in random order. The teacher model is responsible for providing the training data, which in this case means it must determine which instances (subset) of the mini-batch of data will be fed to the student. In order to reach the convergence faster, the reward was set to relate to the speed the student model learns. <br />
<br />
The authors also designed a state feature vector <math> g(s) </math> in order to efficiently represent the current states which include arrived training data and the student model. Within the State Features, there are three categories including Data features, student model features and the combination of both data and learner model. This state feature will be computed when each mini-batch of data arrives.<br />
<br />
<br />
The optimizer for training the teacher model is the maximum expected reward: <br />
<br />
\begin{align} <br />
J(θ) = E_{φ_θ(a|s)}[R(s,a)]<br />
\end{align}<br />
<br />
Which is non-differentiable w.r.t. <math> θ </math>, thus a likelihood ratio policy gradient algorithm is used to optimize <math> J(θ) </math> (Williams, 1992) [4]<br />
<br />
==Experiments==<br />
<br />
The L2T framework is tested on the following student models: multi-layer perceptron (MLP), ResNet (CNN), and Long-Short-Term-Memory network (RNN). <br />
<br />
The student tasks are Image classification for MNIST, for CIFAR-10, and sentiment classification for IMDB movie review dataset. <br />
<br />
The strategy will be benchmarked against the following teaching strategies:<br />
<br />
::'''NoTeach''': NoTeach removes the entire Teacher-Student paradigm and reverts back to the classical machine learning paradigm. In the context of data teaching, we consider the architecture fixed, and feed data in a pre-determined way. One would pre-define batch-size and cross-validation procedures as needed.<br />
::'''Self-Paced Learning (SPL)''': Teaching by ''hardness'' of data, defined as the loss. This strategy begins by filtering out data with larger loss value to train the student with "easy" data and gradually increases the hardness.<br />
::'''L2T''': The Learning to Teach framework.<br />
::'''RandTeach''': Randomly filter data in each epoch according to the logged ratio of filtered data instances per epoch (as opposed to deliberate and dynamic filtering by L2T).<br />
<br />
<br />
===Training a New Student===<br />
<br />
In the first set of experiments, the datasets or divided into two folds. The first folder is used to train the teacher; This is done by having the teacher train a student network on that half of the data, with a certain portion being used for computing rewards. After training, the teacher parameters are fixed, and used to train a new student network (with the same structure) on the second half of the dataset. When teaching a new student with the same model architecture, we observe that L2T achieves significantly faster convergence than other strategies across all tasks, especially compared to the NoTeach and RandTeach methods:<br />
<br />
[[File: L2T_speed.png | 1100px|center]]<br />
<br />
===Filtration Number===<br />
<br />
When investigating the details of filtered data instances per epoch, for the two image classification tasks, the L2T teacher filters an increasing amount of data as training goes on. The authors' intuition for the two image classification tasks is that the student model can learn from harder instances of data from the beginning, and thus the teacher can filter redundant data. In contrast, for training while for the natural language task, the student model must first learn from easy data instances.<br />
<br />
[[File: L2T_fig3.png | 1100px|center]]<br />
<br />
===Teaching New Student with Different Model Architecture===<br />
<br />
In this part, first a teacher model is trained by interacting with a student model. Then using the teacher model, another student model<br />
which has a different model architecture is taught.<br />
The results of Applying the teacher trained on ResNet32 to teach other architectures is shown below. The L2T algorithm can be seen to obtain higher accuracies earlier than the SPL, RandTeach, or NoTeach algorithms.<br />
<br />
[[File: L2T_fig4.png | 1100px|center]]<br />
<br />
===Training Time Analysis===<br />
<br />
The learning curves demonstrate the efficiency in accuracy achieved by the L2T over the other strategies. This is especially evident during the earlier training stages.<br />
<br />
[[File: L2T_fig5.png | 600px|center]]<br />
<br />
===Accuracy Improvement===<br />
<br />
When comparing training accuracy on the IMDB sentiment classification task, L2T improves on teaching policy over NoTeach and SPL.<br />
<br />
[[File: L2T_t1.png | 500px|center]]<br />
<br />
Table 1 shows that we boost the convergence speed, while the teacher model improves final accuracy. The student model is the LSTM network trained on IMDB. Prior to teaching the student model, we train the teacher model on half of the training data, and define the terminal reward as the set accuracy after the teacher model trains the student for 15 epochs. Then the teacher model is applied to train the student model on the full dataset till its convergence. The state features are kept the same as those in previous experiments. We can see that L2T achieves better classification accuracy for training LSTM network, surpassing the SPL baseline by more than 0.6 point (with p value < 0.001).<br />
<br />
=Future Work=<br />
<br />
There is some useful future work that can be extended from this work: <br />
<br />
1) Recent advances in multi-agent reinforcement learning could be tried on the Reinforcement Learning problem formulation of this paper. <br />
<br />
2) Some human in the loop architectures like CHAT and HAT (https://www.ijcai.org/proceedings/2017/0422.pdf) should give better results for the same framework. <br />
<br />
3) It would be interesting to try out the framework suggested in this paper (L2T) in Imperfect information and partially observable settings. <br />
<br />
4) As they have focused on data teaching exploring loss function teaching would be interesting.<br />
<br />
=Critique=<br />
<br />
While the conceptual framework of L2T is sound, the paper only experimentally demonstrates efficacy for ''data teaching'' which would seem to be the simplest to implement. The feasibility and effectiveness of teaching the loss function and hypothesis space are not explored in a real-world scenario. Furthermore, the experimental results for data teaching suggest that the speed of convergence is the main improvement over other teaching strategies whereas the difference in accuracy less remarkable. The paper also assesses accuracy only by comparing L2T with NoTeach and SPL on the IMDB classification task, the improvement (or lack thereof) on the other classification tasks and teaching strategies is omitted. Again, this distinction is not possible to assess in loss function or hypothesis space teaching within the scope of this paper. They could have included larger datasets such as ImageNet and CIFAR100 in their experiments which would have provided some more insight.<br />
<br />
The idea of having a generalizable teacher model to enhance student learning is admirable. In fact, the L2T framework is similar to the reinforcement learning actor-critic model, which is known to be effective. In general, one expects an effective teacher model would facilitate transfer learning and can significantly reduce student model training time. However, the T2L framework seems to fall short of that goal. Consider the CIFAR10 training scenario, the L2T model achieve 85% accuracy after 2 million training data, which is only about 3% more accuracy than a no-teacher model. Perhaps in the future, the L2T framework can improve and produce better performance.</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Learning_to_Teach&diff=40747Learning to Teach2018-11-22T02:27:51Z<p>Gchalato: /* Experiments */</p>
<hr />
<div><br />
<br />
=Introduction=<br />
<br />
This paper proposed the "learning to teach" (L2T) framework with two intelligent agents: a student model/agent, corresponding to the learner in traditional machine learning algorithms, and a teacher model/agent, determining the appropriate data, loss function, and hypothesis space to facilitate the learning of the student model.<br />
<br />
In modern human society, the role of teaching is heavily implicated in our education system; the goal is to equip students with the necessary knowledge and skills in an efficient manner. This is the fundamental ''student'' and ''teacher'' framework on which education stands. However, in the field of artificial intelligence (AI) and specifically machine learning, researchers have focused most of their efforts on the ''student'' (ie. designing various optimization algorithms to enhance the learning ability of intelligent agents). The paper argues that a formal study on the role of ‘teaching’ in AI is required. Analogous to teaching in human society, the teaching framework can: select training data that corresponds to the appropriate teaching materials (e.g. textbooks selected for the right difficulty), design loss functions that correspond to targeted examinations, and define the hypothesis space that corresponds to imparting the proper methodologies. Furthermore, an optimization framework (instead of heuristics) should be used to update the teaching skills based on the feedback from students, so as to achieve teacher-student co-evolution.<br />
<br />
Thus, the training phase of L2T would have several episodes of interactions between the teacher and the student model. Based on the state information in each step, the teacher model would update the teaching actions so that the student model could perform better on the Machine Learning problem. The student model would then provide reward signals back to the teacher model. These reward signals are used by the teacher model as part of the Reinforcement Learning process to update its parameters. This process is end-to-end trainable and the authors are convinced that once converged, the teacher model could be applied to new learning scenarios and even new students, without extra efforts on re-training.<br />
<br />
To demonstrate the practical value of the proposed approach, the '''training data scheduling''' problem is chosen as an example. The authors show that by using the proposed method to adaptively select the most<br />
suitable training data, they can significantly improve the accuracy and convergence speed of various neural networks including multi-layer perceptron (MLP), convolutional neural networks (CNNs)<br />
and recurrent neural networks (RNNs), for different applications including image classification and text understanding.<br />
<br />
=Related Work=<br />
The L2T framework connects with two emerging trends in machine learning. The first is the movement from simple to advanced learning. This includes meta-learning (Schmidhuber, 1987; Thrun & Pratt, 2012) which explores automatic learning by transferring learned knowledge from meta tasks [1]. This approach has been applied to few-shot learning scenarios and in designing general optimizers and neural network architectures. (Hochreiter et al., 2001; Andrychowicz et al., 2016; Li & Malik, 2016; Zoph & Le, 2017)<br />
<br />
The second is the teaching, which can be classified into either machine-teaching (Zhu, 2015) [2] or hardness based methods. The former seeks to construct a minimal training set for the student to learn a target model (ie. an oracle). The latter assumes an order of data from easy instances to hard ones, hardness being determined in different ways. In curriculum learning (CL) (Bengio et al, 2009; Spitkovsky et al. 2010; Tsvetkov et al, 2016) [3] measures hardness through heuristics of the data while self-paced learning (SPL) (Kumar et al., 2010; Lee & Grauman, 2011; Jiang et al., 2014; Supancic & Ramanan, 2013) [4] measures hardness by loss on data. <br />
<br />
The limitations of these works include the lack of a formally defined teaching problem, and the reliance on heuristics and fixed rules, which hinders generalization of the teaching task.<br />
<br />
=Learning to Teach=<br />
To introduce the problem and framework, without loss of generality, consider the setting of supervised learning.<br />
<br />
In supervised learning, each sample <math>x</math> is from a fixed but unknown distribution <math>P(x)</math>, and the corresponding label <math> y </math> is from a fixed but unknown distribution <math>P(y|x) </math>. The goal is to find a function <math>f_\omega(x)</math> with parameter vector <math>\omega</math> that minimizes the gap between the predicted label and the actual label.<br />
<br />
<br />
<br />
==Problem Definition==<br />
The student model, denoted &mu;(), takes the set of training data <math> D </math>, the function class <math> Ω </math>, and loss function <math> L </math> as input to output a function, <math> f(ω) </math>, with parameter <math>ω^*</math> which minimizes risk <math>R(ω)</math> as in:<br />
<br />
\begin{align*}<br />
ω^* = arg min_{w \in \Omega} \sum_{x,y \in D} L(y, f_ω(x)) =: \mu (D, L, \Omega)<br />
\end{align*}<br />
<br />
The teaching model, denoted φ, tries to provide <math> D </math>, <math> L </math>, and <math> Ω </math> (or any combination, denoted <math> A </math>) to the student model such that the student model either achieves lower risk R(ω) or progresses as fast as possible.<br />
<br />
::'''Training Data''': Outputting a good training set <math> D </math>, analogous to human teachers providing students with proper learning materials such as textbooks.<br />
::'''Loss Function''': Designing a good loss function <math> L </math> , analogous to providing useful assessment criteria for students.<br />
::'''Hypothesis Space''': Defining a good function class <math> Ω </math> which the student model can select from. This is analogous to human teachers providing appropriate context, eg. middle school students taught math with basic algebra while undergraduate students are taught with calculus. Different Ω leads to different errors and optimization problem (Mohri et al., 2012).<br />
<br />
==Framework==<br />
The training phase consists of the teacher providing the student with the subset <math> A_{train} </math> of <math> A </math> and then taking feedback to improve its own parameters. The L2T process is outlined in figure below:<br />
<br />
[[File: L2T_process.png | 500px|center]]<br />
<br />
* <math> s_t &isin; S </math> represents information available to the teacher model at time <math> t </math>. <math> s_t </math> is typically constructed from the current student model <math> f_{t−1} </math> and the past teaching history of the teacher model. <math> S </math> represents the set of states.<br />
* <math> a_t &isin; A </math> represents action taken the teacher model at time <math> t </math>, given state <math>s_t</math>. <math> A </math> represents the set of actions, where the action(s) can be any combination of teaching tasks involving the training data, loss function, and hypothesis space. <br />
* <math> φ_θ : S → A </math> is policy used by the teacher model to generate its action <math> φ_θ(s_t) = a_t </math><br />
* Student model takes <math> a_t </math> as input and outputs function <math> f_t </math>, by using the conventional ML techniques.<br />
<br />
Once the training process converges, the teacher model may be utilized to teach a different subset of <math> A </math> or teach a different student model.<br />
<br />
=Application=<br />
<br />
There are different approaches to training the teacher model, this paper will apply reinforcement learning with <math> φ_θ </math> being the ''policy'' that interacts with <math> S </math>, the ''environment''. The paper applies data teaching to train a deep neural network student, <math> f </math>, for several classification tasks. Thus the student feedback measure will be classification accuracy. Its learning rule will be mini-batch stochastic gradient descent, where batches of data will arrive sequentially in random order. The teacher model is responsible for providing the training data, which in this case means it must determine which instances (subset) of the mini-batch of data will be fed to the student. In order to reach the convergence faster, the reward was set to relate to the speed the student model learns. <br />
<br />
The authors also designed a state feature vector <math> g(s) </math> in order to efficiently represent the current states which include arrived training data and the student model. Within the State Features, there are three categories including Data features, student model features and the combination of both data and learner model. This state feature will be computed when each mini-batch of data arrives.<br />
<br />
<br />
The optimizer for training the teacher model is the maximum expected reward: <br />
<br />
\begin{align} <br />
J(θ) = E_{φ_θ(a|s)}[R(s,a)]<br />
\end{align}<br />
<br />
Which is non-differentiable w.r.t. <math> θ </math>, thus a likelihood ratio policy gradient algorithm is used to optimize <math> J(θ) </math> (Williams, 1992) [4]<br />
<br />
==Experiments==<br />
<br />
The L2T framework is tested on the following student models: multi-layer perceptron (MLP), ResNet (CNN), and Long-Short-Term-Memory network (RNN). <br />
<br />
The student tasks are Image classification for MNIST, for CIFAR-10, and sentiment classification for IMDB movie review dataset. <br />
<br />
The strategy will be benchmarked against the following teaching strategies:<br />
<br />
::'''NoTeach''': NoTeach removes the entire Teacher-Student Paradigm and reverts back to the classical machine learning paradigm. In the context of data teaching, we consider the architecture fixed, and feed data in a pre-determined way. One would pre-define batch-size and cross-validation procedures as needed.<br />
::'''Self-Paced Learning (SPL)''': Teaching by ''hardness'' of data, defined as the loss. This strategy begins by filtering out data with larger loss value to train the student with "easy" data and gradually increases the hardness.<br />
::'''L2T''': The Learning to Teach framework.<br />
::'''RandTeach''': Randomly filter data in each epoch according to the logged ratio of filtered data instances per epoch (as opposed to deliberate and dynamic filtering by L2T).<br />
<br />
<br />
===Training a New Student===<br />
<br />
In the first set of experiments, the datasets or divided into two folds. The first folder is used to train the teacher; This is done by having the teacher train a student network on that half of the data, with a certain portion being used for computing rewards. After training, the teacher parameters are fixed, and used to train a new student network (with the same structure) on the second half of the dataset. When teaching a new student with the same model architecture, we observe that L2T achieves significantly faster convergence than other strategies across all tasks, especially compared to the NoTeach and RandTeach methods:<br />
<br />
[[File: L2T_speed.png | 1100px|center]]<br />
<br />
===Filtration Number===<br />
<br />
When investigating the details of filtered data instances per epoch, for the two image classification tasks, the L2T teacher filters an increasing amount of data as training goes on. The authors' intuition for the two image classification tasks is that the student model can learn from harder instances of data from the beginning, and thus the teacher can filter redundant data. In contrast, for training while for the natural language task, the student model must first learn from easy data instances.<br />
<br />
[[File: L2T_fig3.png | 1100px|center]]<br />
<br />
===Teaching New Student with Different Model Architecture===<br />
<br />
In this part, first a teacher model is trained by interacting with a student model. Then using the teacher model, another student model<br />
which has a different model architecture is taught.<br />
The results of Applying the teacher trained on ResNet32 to teach other architectures is shown below. The L2T algorithm can be seen to obtain higher accuracies earlier than the SPL, RandTeach, or NoTeach algorithms.<br />
<br />
[[File: L2T_fig4.png | 1100px|center]]<br />
<br />
===Training Time Analysis===<br />
<br />
The learning curves demonstrate the efficiency in accuracy achieved by the L2T over the other strategies. This is especially evident during the earlier training stages.<br />
<br />
[[File: L2T_fig5.png | 600px|center]]<br />
<br />
===Accuracy Improvement===<br />
<br />
When comparing training accuracy on the IMDB sentiment classification task, L2T improves on teaching policy over NoTeach and SPL.<br />
<br />
[[File: L2T_t1.png | 500px|center]]<br />
<br />
Table 1 shows that we boost the convergence speed, while the teacher model improves final accuracy. The student model is the LSTM network trained on IMDB. Prior to teaching the student model, we train the teacher model on half of the training data, and define the terminal reward as the set accuracy after the teacher model trains the student for 15 epochs. Then the teacher model is applied to train the student model on the full dataset till its convergence. The state features are kept the same as those in previous experiments. We can see that L2T achieves better classification accuracy for training LSTM network, surpassing the SPL baseline by more than 0.6 point (with p value < 0.001).<br />
<br />
=Future Work=<br />
<br />
There is some useful future work that can be extended from this work: <br />
<br />
1) Recent advances in multi-agent reinforcement learning could be tried on the Reinforcement Learning problem formulation of this paper. <br />
<br />
2) Some human in the loop architectures like CHAT and HAT (https://www.ijcai.org/proceedings/2017/0422.pdf) should give better results for the same framework. <br />
<br />
3) It would be interesting to try out the framework suggested in this paper (L2T) in Imperfect information and partially observable settings. <br />
<br />
4) As they have focused on data teaching exploring loss function teaching would be interesting.<br />
<br />
=Critique=<br />
<br />
While the conceptual framework of L2T is sound, the paper only experimentally demonstrates efficacy for ''data teaching'' which would seem to be the simplest to implement. The feasibility and effectiveness of teaching the loss function and hypothesis space are not explored in a real-world scenario. Furthermore, the experimental results for data teaching suggest that the speed of convergence is the main improvement over other teaching strategies whereas the difference in accuracy less remarkable. The paper also assesses accuracy only by comparing L2T with NoTeach and SPL on the IMDB classification task, the improvement (or lack thereof) on the other classification tasks and teaching strategies is omitted. Again, this distinction is not possible to assess in loss function or hypothesis space teaching within the scope of this paper. They could have included larger datasets such as ImageNet and CIFAR100 in their experiments which would have provided some more insight.<br />
<br />
The idea of having a generalizable teacher model to enhance student learning is admirable. In fact, the L2T framework is similar to the reinforcement learning actor-critic model, which is known to be effective. In general, one expects an effective teacher model would facilitate transfer learning and can significantly reduce student model training time. However, the T2L framework seems to fall short of that goal. Consider the CIFAR10 training scenario, the L2T model achieve 85% accuracy after 2 million training data, which is only about 3% more accuracy than a no-teacher model. Perhaps in the future, the L2T framework can improve and produce better performance.</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=stat946F18/differentiableplasticity&diff=40739stat946F18/differentiableplasticity2018-11-22T01:37:01Z<p>Gchalato: /* Related Work */</p>
<hr />
<div>'''Differentiable Plasticity: ''' Summary of the ICML 2018 paper https://arxiv.org/abs/1804.02464<br />
<br />
= Presented by =<br />
<br />
1. Ganapathi Subramanian, Sriram [Quest ID: 20676799]<br />
<br />
= Motivation =<br />
Machine Learning models often employ extensive training over massive dataset of training examples in order to learn a single complex task very well. However, biological agents contrast this learning style by exhibiting a remarkable ability to learn quickly and efficiently from ongoing experience. <br />
<br />
1. Neural Networks naturally have a static architecture. Once a Neural Network is trained, the network architecture components (ex. network connections) cannot be changed and effectively, learning stops with the training step. If a different task needs to be considered, then the agent must be trained again from scratch. <br />
<br />
2. Plasticity is the characteristic of biological systems present in humans, which can change network connections over time. For instance, animals can learn to navigate and remember the location and optimal path to food sources. This enables lifelong learning in biological systems and thus, allows for adaptation to dynamic changes in the environment with great sample efficiency in the data observed. This is called synaptic plasticity, which is based on the Hebb's rule (i.e. if a neuron repeatedly takes part in making another neuron fire, the connection between them is strengthened). Neural networks are very far from achieving synaptic plasticity. <br />
<br />
3. Differentiable plasticity is a step in this direction. The behavior of the plastic connection is trained using gradient descent so that the previously trained networks can adapt to changing conditions thus mimicking dynamic learning of rewarding or detrimental behaviour.<br />
<br />
Example: Using the current state of the art supervised learning examples, we can train Neural Networks to recognize specific letters that it has seen during training. Using lifelong learning, the agent can develop a knowledge about any alphabet, including those that it has never been exposed to during training.<br />
<br />
= Objectives =<br />
The paper has the following objectives: <br />
<br />
1. To tackle the problem of meta-learning (learning to learn). <br />
<br />
2. To design neural networks with plastic connections with a special emphasis on gradient descent capability for backpropagation training. <br />
<br />
3. To use backpropagation to optimize both the base weights and the amount of plasticity in each connection. <br />
<br />
4. To demonstrate the performance of such networks on three complex and different domains, namely complex pattern memorization, one shot classification and reinforcement learning.<br />
<br />
= Important Terms =<br />
<br />
Hebb’s rule: This is a famous rule in neuroscience. It defines the relationship of activities between neurons with their connection. It states that if a neuron repeatedly takes part in making another neuron fire, the connection between them is strengthened. Also summarized as "neurons that fire together, wire together".<br />
<br />
= Related Work =<br />
<br />
Previous Approaches to solving this problem are summarized below: <br />
<br />
1. Train standard recurrent neural networks to incorporate past experience in their future responses within each episode. For the learning abilities, the RNN is attached with an external content-addressable memory bank. An attention mechanism within the controller network does the read-write to the memory bank and thus enables fast memorization. <br />
<br />
2. Augment each weight with a plastic component that automatically grows and decays as a function of inputs and outputs. All connection have the same non-trainable plasticity and only the corresponding weights are trained. Recent approaches have tried fast-weights which augments recurrent networks with fast-changing Hebbian weights and computes the activation function at each step. The network has a high bias towards the recently seen patterns. <br />
<br />
3. Optimize the learning rule itself, instead of the connections. A parametrized learning rule is used where the structure of the network is fixed beforehand. <br />
<br />
4. Have all the weight updates to be computed on the fly by the network itself or by a separate network at each time step. Pros are the flexibility and the cons are the large learning burden placed on the network. <br />
<br />
5. Perform gradient descent via propagation during the episode. The meta-learning involves training the base network for it to be fine-tuned using additional gradient descent. <br />
<br />
6. For classification tasks, the idea of learning a “new object” is analogous to understanding how the embedding of a test example relates to the embeddings of classes known in the test set. Specifically, once we have embeddings to represent a particular class, given new data, we simply extract the embedding of the test sample and connect it to an embedding with a known class (through whichever distance metric we decide to use). Note however, this does not actually “learn-to-learn”, in that the process of prediction never changes. Embeddings are always held constant, unless the test cases, when classified, are used to redefine the prototypical embedding of a class.<br />
<br />
The superiority of the trainable synaptic plasticity for the meta-learning approach are as follows: <br />
<br />
1. Great potential for flexibility. Example, Memory Networks enforce a specific memory storage model in which memories must be embedded in fixed-size vectors and retrieved through some attention mechanism. In contrast, trainable synaptic plasticity translates into very different forms of memory, the exact implementation of which can be determined<br />
by (trainable) network structure.<br />
<br />
2. Fixed-weight recurrent networks, meanwhile, require neurons to be used for both<br />
storage and computation which increases the computational burdens on neurons. This is avoided in the approach suggested in the paper. <br />
<br />
3. Non-trainable plasticity networks can exploit network connectivity for storage of short-term information, but their uniform, non-trainable plasticity imposes a stereotypical behavior on these memories. In the synaptic plasticity, the amount and rate of plasticity are actively molded by the mechanism itself. Also, it allows for more sustained memory.<br />
<br />
= Model =<br />
<br />
The formulation proposed in the paper is in such a way that the plastic and non-plastic components for each connection are kept separate, while multiple Hebbian rules can be easily defined. <br />
<br />
Model Components: <br />
<br />
1. A connection between any two neurons <math display = "inline">i</math> and <math display = "inline">j</math> has both a fixed component and a plastic component. <br />
<br />
2. The fixed part is just a traditional connection weight, <math display = "inline">w_{i,j}</math> . The plastic part is stored in a Hebbian trace, <math display = "inline">H_{i,j}</math>, which varies during a<br />
lifetime according to ongoing inputs and outputs.<br />
<br />
3. The relative importance of plastic and fixed components in the connection is structurally determined by the plasticity<br />
coefficient, <math display = "inline">\alpha_{i,j}</math>, which multiplies the Hebbian trace to form<br />
the full plastic component of the connection. <br />
<br />
The network equations for the output <math display = "inline">x_j(t)</math> of the neuron <math display = "inline">j</math> are as follows: <br />
<br />
<br />
<math display="block"><br />
x_j(t) = \sigma \Big\{\displaystyle \sum_{i \in ~\text{inputs}}[w_{i,j}x_i(t-1) + \alpha_{i,j} H_{i,j}(t)x_i(t-1)] \Big\}<br />
</math><br />
<br />
<br />
<br />
<math display="block"><br />
H_{i,j}(t+1) = \eta x_i(t-1) x_j(t) + (1 - \eta) H_{i,j}(t) <br />
</math><br />
<br />
Here the first equation gives the activation function, where the <math display = "inline">w_{i,j}</math> is a fixed component and the remaining term (<math display = "inline"> \alpha_{i,j} H_{i,j}(t))x_i(t-1) </math>) is a plastic component. The <math display = "inline">\sigma</math> is a nonlinear function, chosen to be tanh in this paper. The <math display = "inline">H_{i,j}</math> in the second equation is updated as a function of ongoing inputs and outputs after being initialized to zero at each episode. In contrast, <math display = "inline">w_{i,j}</math> and <math display = "inline">\alpha_{i,j}</math> are the structural parameters trained by gradient descent and conserved across episodes.<br />
<br />
From the first equation above, a connection is fully fixed if <math display = "inline">\alpha = 0 </math>. Alternatively, a connection is fully plastic if <math display = "inline">w = 0</math>. Otherwise, the connection has both a fixed and plastic components. <br />
<br />
<br />
The <math display = "inline">\eta</math> denotes the learning rate, which is also an optimized parameter of the network. After this training, the agent can learn automatically from ongoing experience. In equation 2, the <math display = "inline">\eta</math> could make the Hebbian traces decay to 0 in the absence of input. This leads to the following form of the equation as follows: <br />
<br />
<br />
<math display="block"><br />
H_{i,j}(t+1) = H_{i,j}(t) + \eta x_j(t)(x_i(t-1) - x_j(t)H_{i,j}(t))<br />
</math><br />
<br />
The Hebbian trace is a representation of concurrent firing of <math>x_j, x_i</math> over past time-steps , and is meant to strengthen connection between neurons that are often activated together.<br />
<br />
= Experiment 1 - Binary Pattern Memorization =<br />
<br />
<br />
<br />
This test involves quickly memorizing sets of arbitrary high-dimensional patterns and reconstructing the same while being exposed to partial, degraded versions of them. This is a very simple test as it is already known that hand designed recurrent networks with a Hebbian plastic connection can already solve it for binary patterns.<br />
<br />
<br />
<br />
[[File:binarypatternrecog.png | 650px|thumb|center|Figure 1: Pattern Memorization experiment - Input Structure and Architecture]]<br />
<br />
<br />
<br />
'''Steps in the experiment:''' <br />
<br />
1) The network is a set of five binary patterns in succession as shown in the figure 1. Each of these patterns has 1,000 elements, for which each element is binary-valued (1 or -1). Here, dark red corresponds to the value 1, and dark blue corresponds to the value -1. <br />
<br />
2) The few shot learning paradigm is followed, where each pattern is shown for 10-time steps, with 3-time steps of zero input between the presentations and the whole sequence of patterns is presented 3 times in random order. <br />
<br />
3) One of the presented patterns is chosen in random order and degraded by setting half of its bits to 0. <br />
<br />
4) This degraded pattern is then fed to the network. The network has to reproduce the correct full pattern in its output using its memory that it developed during training. <br />
<br />
<br />
'''The architecture of the network is described as follows:''' <br />
<br />
1) It is a fully connected RNN with one neuron per pattern element, plus one fixed-output neuron (bias). There are a total of 1,001 neurons. <br />
<br />
2) Value of each neuron is clamped to the value of the corresponding element in the pattern if the value is not 0. If the value is 0, the corresponding neurons do not receive pattern input and must use what it gets from lateral connections and reconstruct the correct, expected output values. <br />
<br />
3) Outputs are read from the activation of the neurons. <br />
<br />
4) The performance evaluation is done by computing the loss between the final network output and the correct expected pattern. <br />
<br />
5) The gradient of the error over the <math display = "inline">w_{i,j}</math> and the <math display = "inline">\alpha_{i,j}</math> coefficients is computed by backpropagation and optimized through Adam solver with learning rate 0.001. <br />
<br />
6) The simple decaying Hebbian formula in Equation 2 is used to update the Hebbian traces. Each network has 2 trainable parameters <math display = "inline">w</math> and <math display = "inline">\alpha</math> for each connection, thus there are a total 1,001 <math display = "inline">\times</math> 1,001 <math display = "inline">\times</math> 2 = 2,004,002 trainable parameters. <br />
<br />
[[File:exp1results.png | 650px|thumb|center|Figure 2:Experiment 1 - Pattern Memorization Results]]<br />
<br />
<br />
The results are shown in the figure 2 where 10 runs are considered. The error becomes quite low after about 200 episodes of training. <br />
<br />
[[File:exp1nonplasticresults.png| 650px|thumb|center|Figure 3: Pattern Memorization results with non plastic networks]]<br />
<br />
<br />
<br />
'''Comparison with Non-Plastic Networks:''' <br />
<br />
1) Non-plastic networks can solve this task but require additional neurons to solve this task in principle. In practice, the authors say that the task is not solved using Non-plastic RNN or LSTM. <br />
<br />
2) The figure 3 shows the results using non-plastic networks. The best results required the addition of 2000 extra neurons. <br />
<br />
3) For non-plastic RNN, the error flattens around 0.13 which is quite high. Using LSTMs, the task can be solved albeit imperfectly and also the error rate reduces drastically t0 around 0.001. <br />
<br />
4) The plastic network solves the task very quickly with the mean error going below 0.01 within 2000 episodes which are mentioned to be 250 times faster than the LSTM.<br />
<br />
= Experiment 2 - Memorizing network images=<br />
<br />
This task is an image reconstruction task that where a network is trained on a set of natural images which it looks to memorize. The natural images with graded pixel values contain more information per element as compared to the last experiment. So this experiment is inherently more complex than the previous ones. Then one image is chosen at random and half the image is displayed to the agent. The task is to complete the image. The paper shows that this method effectively solves this task which other state-of-the-art network architectures fail to solve. <br />
<br />
The experiment is as follows: <br />
<br />
1) Images are from the CIFAR-10 database where there are a total of 60000 images each of size 32 <math display = "inline">\times</math> 32. <br />
<br />
2) The architecture has 1025 neurons in total with a total of 2 <math display = "inline">\times</math> 1025 <math display = "inline">\times</math> 1025 = 2101250 parameters. <br />
<br />
3) Each episode has 3 pictures, shown 3 times for 20-time steps each time, with 3-time steps of zero input between the presentations. <br />
<br />
4) The images are degraded by zeroing out one full contiguous half of the image to prevent a trivial solution of simply reconstructing the missing pixel as the average of its neighbors.<br />
<br />
[[File:exp2results.png| 650px|thumb|center|Figure 4: Natural Image memorization results]]<br />
<br />
<br />
<br />
The results are shown in figure 4. The final output of the network is shown in the last column which is the reconstructed image. The results show that the model has learned to perform this task. <br />
<br />
[[File:exp2weights.png| 650px|thumb|center|Figure 5: Final matrices and plasticity coefficients]]<br />
<br />
The final weight matrix and plasticity coefficients matrix are shown in the figure 5. The plasticity matrix shows a structure related to the high correlation of neighboring pixels and half-field zeroing in test images. <br />
<br />
The full plastic network is compared against a similar architecture with shared plasticity coefficients, where all connections share the same <math display = "inline">\alpha</math> value. So, the single parameter is shared across all connections is trained. <br />
<br />
[[File:independentvsshared.png| 650px|thumb|center|Figure 6: Comparing independent and shared <math display = "inline">\alpha</math> value runs]]<br />
<br />
The figure 6 shows the result of comparison where the independent plasticity coefficient for each connection has better performances. Thus the structure observed in the weight matrices of the results is actually useful.<br />
<br />
<br />
= Experiment 3 - Omniglot task =<br />
<br />
This task involves handwritten symbol recognition. It is a standard task for one-shot and few-shot learning. <br />
<br />
===Experimental Setup: ===<br />
<br />
1) The Omniglot data set is a collection of handwritten characters from various writing systems, including 20 instances each of 1,623 different handwritten characters, written by different subjects.<br />
<br />
[[File:Omniglot Dataset.JPG|400px|center]]<br />
<br />
2) In each episode, N character classes are randomly selected and K instances from each class are sampled. <br />
<br />
3) These instances, together with the class label (from 1 to N), are shown to the model. <br />
<br />
4) Then, a new, unlabeled instance is sampled from one of the N classes and shown to the model.<br />
<br />
5) Model performance is defined as the model’s accuracy in classifying this unlabeled example.<br />
<br />
===Architecture: ===<br />
<br />
1) Model architecture has 4 convolutional layers with 3 <math display = "inline">\times</math> 3 receptive fields and 64 channels. <br />
<br />
2) All convolutions have a stride of 2 to reduce the dimensionality between layers. <br />
<br />
3) The output is a single vector of 64 features, which feeds into an N-way softmax. <br />
<br />
4) The label of the current character is also concurrently fed as a one-hot encoding to this softmax layer, to serve as a guide for the correct output when a label is present.<br />
<br />
===Plasticity in the architecture: ===<br />
<br />
1) Plasticity is applied to the weights from the final layer to the softmax layer, leaving the rest of the convolutional embedding non- plastic. <br />
<br />
2) The expectation is that the convolutional architecture will learn an adequate discriminant between arbitrary handwritten characters and the plastic weights learns to memorize associations between observed patterns and outputs. <br />
<br />
===Data Preparation: ===<br />
<br />
1) The dataset is augmented with rotations by multiples of <math display = "inline">90</math> degrees. <br />
<br />
2) It is divided into 1,523 classes for training and 100 classes (together with their augmentations) for testing. <br />
<br />
3) The networks are trained with an Adam optimizer with a learning rate 3 <math display = "inline">\times 10^{-5}</math>, multiplied by 2/3 every 1M episodes over 5,000,000 episodes. <br />
<br />
4) To evaluate final model performance, 10 models are trained with different random seeds and each of those is tested on 100 episodes using previously unseen test classes.<br />
<br />
===Results: ===<br />
<br />
1) The overall accuracy (i.e. the proportion of episodes with correct classification, aggregated over all test episodes of all runs) is 98.3%, with a 95% confidence interval of 0.80%.<br />
<br />
2) The median accuracy across the 10 runs was 98.5%, indicating consistency in learning.<br />
<br />
{| class="wikitable"<br />
|-<br />
! Memory Networks<br />
! Matching Networks<br />
! ProtoNets<br />
! Memory Module<br />
! MAML<br />
! SNAIL<br />
! DP(This paper)<br />
|-<br />
| 82.8%<br />
| 98.1%<br />
| 97.4%<br />
| 98.4%<br />
| 98.7% <math display = "inline">\pm</math> 0.4<br />
| 99.07% <math display = "inline">\pm</math> 0.16<br />
| 98.03% <math display = "inline">\pm</math> 0.80<br />
|}<br />
<br />
<br />
<br />
3) The above table shows the comparative performance across other non-plastic approaches. The results of the plastic approach are largely similar to those reported for the computationally intensive MAML method and the classification-specialized Matching Networks method. <br />
<br />
4) The performances are slightly below those reported for the SNAIL method, which trains a whole additional temporal-convolution network on top of the convolutional architecture thus having many more parameters.<br />
<br />
5) The conclusion is that a few plastic connections to the output of the network allow for competitive one-shot learning over arbitrary man-made visual symbols.<br />
<br />
= Experiment 4 - Reinforcement learning Maze navigation task =<br />
<br />
This is a maze exploration task where the goal is to teach an agent to reach a goal. The plastic networks are shown to outperform non-plastic ones. <br />
<br />
Experimental setup: <br />
<br />
1) The maze is composed of 9 <math display = "inline">\times</math> 9 squares, surrounded by walls, in which every other square (in either direction) is occupied by a wall. <br />
<br />
[[File:exp4maze.png| 650px|thumb|center|Figure 7: Maze Environment]]<br />
<br />
<br />
2) The maze contains 16 wall square arranged in a regular grid as shown in the figure 7. <br />
<br />
3) At each episode, one non-wall square is randomly chosen as the reward location. When the agent hits this location, it receives a large reward (10.0) and is immediately transported to a random location in the maze Also a small negative reward of -0.1 is provided every time the agent tries to walk into a wall).<br />
<br />
4) Each episode lasts 250-time steps, during which the agent must accumulate as much reward as possible. The reward location is fixed within an episode and randomized across episodes. <br />
<br />
5) The reward is invisible to the agent, and thus the agent only knows it has hit the reward location by the activation of the reward input at the next step.<br />
<br />
6) Inputs to the agent consist of a binary vector describing the 3 <math display = "inline">\times</math> 3 neighborhood centered on the agent (each element being set to 1 or 0 if the corresponding square is or is not a wall), together with the reward at the previous time step. <br />
<br />
7) A2C algorithm is used to meta train the network. <br />
<br />
8) The experiments are run under three conditions: full differentiable plasticity, no plasticity at all, and homogeneous plasticity in which all connections share the same (learnable) <math display = "inline">\alpha</math> parameter. <br />
<br />
9) For each condition, 15 runs with different random seeds are performed. <br />
<br />
<br />
Architecture: <br />
<br />
1) It is a simple recurrent network with 200 neurons, with a softmax layer on top of it to select between the 4 possible actions (up, right, left or down).<br />
<br />
<br />
[[File:exp4performance.png| 650px|thumb|center|Figure 8: Performance curve for the maze navigation experiment]]<br />
<br />
<br />
Results: <br />
<br />
1) The results are shown in the figure 8. The plastic network shows considerably better performance as compared to the other networks.<br />
<br />
2) The non-plastic and homogeneous networks get stuck on a sub-optimal policy. <br />
<br />
3) Thus, the conclusion is that, in this domain, individually sculpting the plasticity of each connection is crucial in reaping the benefits of plasticity for this task.<br />
<br />
= Conclusions =<br />
<br />
<br />
The important contributions from this paper are as follows: <br />
<br />
1) The results show that simple plastic models support efficient meta-learning.<br />
<br />
2) Gradient descent itself is shown to be capable of optimizing the plasticity of a meta-learning system. <br />
<br />
3) The meta-learning is shown to vastly outperform alternative options in the considered experiments. <br />
<br />
4) The method achieved state of the art results on a hard Omniglot test set.<br />
<br />
= Open Source Code =<br />
<br />
Code for this paper can be found at: https://github.com/uber-common/differentiable-plasticity<br />
<br />
<br />
= Critiques =<br />
<br />
The paper addresses an important problem of learning to learn ("meta-learning") and provides a novel framework based on gradient descent to achieve this objective. This paper provides a large scope for future work as many widely used architectures like LSTMs could be tried along with a plastic component. It is also easy to see that the application of such approaches in deep reinforcement learning are also plentiful and there is a good possibility of beating the current baselines in many popular test beds like Atari games using plastic networks. This paper opens up possibilities for a whole class of meta-learning algorithms. <br />
<br />
With regards to the drawbacks of the paper, the paper does not mention how plastic networks will behave if the test sets are completely different from the training dataset. Will the performance be the same as non-plastic networks? It is not very clear if this method will be scalable as there are a large number of parameters to be determined even with the simplest of problems. Also, each experimental domain considered in this paper needed significantly different network architectures (for example in the Omniglot domain plasticity was applied only for the final layers). The paper does not mention any reasons for the specific decisions and if such differences will hold good for other similar problems as well. There has been work in transfer learning applied to both supervised learning and reinforcement learning problems. The authors should have ideally compared plastic networks to performances of some algorithms there as these methods transfer existing knowledge to other related problems and also prevent the need to start training from scratch much similar to the methods adopted in this paper.<br />
<br />
In Experiment 2, the reconstruction of CIFAR-10 images, the authors only provide sample reconstructed images. No quantitative assessment of results is done. It is difficult to judge the generalization of their results. Furthermore, from these results, the authors conclude that their model is good at reconstructing previously unseen images. This claim is quite broad given the relatively simple experiment that was conducted. This is also evident from the network they used, which consisted of only 1000 neurons. Compared with the network in experiment 3, which consisted of a deep 4 layer CNN on a relatively simpler task of classification of Omniglot characters. It would have been more useful if the authors expanded on the image reconstruction task rather than displaying the learned plastic/non-plastic weights. For example, the removed pixels of test images could have been made more random, similar to experiment 1.</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=stat946F18/differentiableplasticity&diff=40738stat946F18/differentiableplasticity2018-11-22T01:36:52Z<p>Gchalato: /* Related Work */</p>
<hr />
<div>'''Differentiable Plasticity: ''' Summary of the ICML 2018 paper https://arxiv.org/abs/1804.02464<br />
<br />
= Presented by =<br />
<br />
1. Ganapathi Subramanian, Sriram [Quest ID: 20676799]<br />
<br />
= Motivation =<br />
Machine Learning models often employ extensive training over massive dataset of training examples in order to learn a single complex task very well. However, biological agents contrast this learning style by exhibiting a remarkable ability to learn quickly and efficiently from ongoing experience. <br />
<br />
1. Neural Networks naturally have a static architecture. Once a Neural Network is trained, the network architecture components (ex. network connections) cannot be changed and effectively, learning stops with the training step. If a different task needs to be considered, then the agent must be trained again from scratch. <br />
<br />
2. Plasticity is the characteristic of biological systems present in humans, which can change network connections over time. For instance, animals can learn to navigate and remember the location and optimal path to food sources. This enables lifelong learning in biological systems and thus, allows for adaptation to dynamic changes in the environment with great sample efficiency in the data observed. This is called synaptic plasticity, which is based on the Hebb's rule (i.e. if a neuron repeatedly takes part in making another neuron fire, the connection between them is strengthened). Neural networks are very far from achieving synaptic plasticity. <br />
<br />
3. Differentiable plasticity is a step in this direction. The behavior of the plastic connection is trained using gradient descent so that the previously trained networks can adapt to changing conditions thus mimicking dynamic learning of rewarding or detrimental behaviour.<br />
<br />
Example: Using the current state of the art supervised learning examples, we can train Neural Networks to recognize specific letters that it has seen during training. Using lifelong learning, the agent can develop a knowledge about any alphabet, including those that it has never been exposed to during training.<br />
<br />
= Objectives =<br />
The paper has the following objectives: <br />
<br />
1. To tackle the problem of meta-learning (learning to learn). <br />
<br />
2. To design neural networks with plastic connections with a special emphasis on gradient descent capability for backpropagation training. <br />
<br />
3. To use backpropagation to optimize both the base weights and the amount of plasticity in each connection. <br />
<br />
4. To demonstrate the performance of such networks on three complex and different domains, namely complex pattern memorization, one shot classification and reinforcement learning.<br />
<br />
= Important Terms =<br />
<br />
Hebb’s rule: This is a famous rule in neuroscience. It defines the relationship of activities between neurons with their connection. It states that if a neuron repeatedly takes part in making another neuron fire, the connection between them is strengthened. Also summarized as "neurons that fire together, wire together".<br />
<br />
= Related Work =<br />
<br />
Previous Approaches to solving this problem are summarized below: <br />
<br />
1. Train standard recurrent neural networks to incorporate past experience in their future responses within each episode. For the learning abilities, the RNN is attached with an external content-addressable memory bank. An attention mechanism within the controller network does the read-write to the memory bank and thus enables fast memorization. <br />
<br />
2. Augment each weight with a plastic component that automatically grows and decays as a function of inputs and outputs. All connection have the same non-trainable plasticity and only the corresponding weights are trained. Recent approaches have tried fast-weights which augments recurrent networks with fast-changing Hebbian weights and computes the activation function at each step. The network has a high bias towards the recently seen patterns. <br />
<br />
3. Optimize the learning rule itself, instead of the connections. A parametrized learning rule is used where the structure of the network is fixed beforehand. <br />
<br />
4. Have all the weight updates to be computed on the fly by the network itself or by a separate network at each time step. Pros are the flexibility and the cons are the large learning burden placed on the network. <br />
<br />
5. Perform gradient descent via propagation during the episode. The meta-learning involves training the base network for it to be fine-tuned using additional gradient descent. <br />
<br />
6. 6. For classification tasks, the idea of learning a “new object” is analogous to understanding how the embedding of a test example relates to the embeddings of classes known in the test set. Specifically, once we have embeddings to represent a particular class, given new data, we simply extract the embedding of the test sample and connect it to an embedding with a known class (through whichever distance metric we decide to use). Note however, this does not actually “learn-to-learn”, in that the process of prediction never changes. Embeddings are always held constant, unless the test cases, when classified, are used to redefine the prototypical embedding of a class.<br />
<br />
The superiority of the trainable synaptic plasticity for the meta-learning approach are as follows: <br />
<br />
1. Great potential for flexibility. Example, Memory Networks enforce a specific memory storage model in which memories must be embedded in fixed-size vectors and retrieved through some attention mechanism. In contrast, trainable synaptic plasticity translates into very different forms of memory, the exact implementation of which can be determined<br />
by (trainable) network structure.<br />
<br />
2. Fixed-weight recurrent networks, meanwhile, require neurons to be used for both<br />
storage and computation which increases the computational burdens on neurons. This is avoided in the approach suggested in the paper. <br />
<br />
3. Non-trainable plasticity networks can exploit network connectivity for storage of short-term information, but their uniform, non-trainable plasticity imposes a stereotypical behavior on these memories. In the synaptic plasticity, the amount and rate of plasticity are actively molded by the mechanism itself. Also, it allows for more sustained memory.<br />
<br />
= Model =<br />
<br />
The formulation proposed in the paper is in such a way that the plastic and non-plastic components for each connection are kept separate, while multiple Hebbian rules can be easily defined. <br />
<br />
Model Components: <br />
<br />
1. A connection between any two neurons <math display = "inline">i</math> and <math display = "inline">j</math> has both a fixed component and a plastic component. <br />
<br />
2. The fixed part is just a traditional connection weight, <math display = "inline">w_{i,j}</math> . The plastic part is stored in a Hebbian trace, <math display = "inline">H_{i,j}</math>, which varies during a<br />
lifetime according to ongoing inputs and outputs.<br />
<br />
3. The relative importance of plastic and fixed components in the connection is structurally determined by the plasticity<br />
coefficient, <math display = "inline">\alpha_{i,j}</math>, which multiplies the Hebbian trace to form<br />
the full plastic component of the connection. <br />
<br />
The network equations for the output <math display = "inline">x_j(t)</math> of the neuron <math display = "inline">j</math> are as follows: <br />
<br />
<br />
<math display="block"><br />
x_j(t) = \sigma \Big\{\displaystyle \sum_{i \in ~\text{inputs}}[w_{i,j}x_i(t-1) + \alpha_{i,j} H_{i,j}(t)x_i(t-1)] \Big\}<br />
</math><br />
<br />
<br />
<br />
<math display="block"><br />
H_{i,j}(t+1) = \eta x_i(t-1) x_j(t) + (1 - \eta) H_{i,j}(t) <br />
</math><br />
<br />
Here the first equation gives the activation function, where the <math display = "inline">w_{i,j}</math> is a fixed component and the remaining term (<math display = "inline"> \alpha_{i,j} H_{i,j}(t))x_i(t-1) </math>) is a plastic component. The <math display = "inline">\sigma</math> is a nonlinear function, chosen to be tanh in this paper. The <math display = "inline">H_{i,j}</math> in the second equation is updated as a function of ongoing inputs and outputs after being initialized to zero at each episode. In contrast, <math display = "inline">w_{i,j}</math> and <math display = "inline">\alpha_{i,j}</math> are the structural parameters trained by gradient descent and conserved across episodes.<br />
<br />
From the first equation above, a connection is fully fixed if <math display = "inline">\alpha = 0 </math>. Alternatively, a connection is fully plastic if <math display = "inline">w = 0</math>. Otherwise, the connection has both a fixed and plastic components. <br />
<br />
<br />
The <math display = "inline">\eta</math> denotes the learning rate, which is also an optimized parameter of the network. After this training, the agent can learn automatically from ongoing experience. In equation 2, the <math display = "inline">\eta</math> could make the Hebbian traces decay to 0 in the absence of input. This leads to the following form of the equation as follows: <br />
<br />
<br />
<math display="block"><br />
H_{i,j}(t+1) = H_{i,j}(t) + \eta x_j(t)(x_i(t-1) - x_j(t)H_{i,j}(t))<br />
</math><br />
<br />
The Hebbian trace is a representation of concurrent firing of <math>x_j, x_i</math> over past time-steps , and is meant to strengthen connection between neurons that are often activated together.<br />
<br />
= Experiment 1 - Binary Pattern Memorization =<br />
<br />
<br />
<br />
This test involves quickly memorizing sets of arbitrary high-dimensional patterns and reconstructing the same while being exposed to partial, degraded versions of them. This is a very simple test as it is already known that hand designed recurrent networks with a Hebbian plastic connection can already solve it for binary patterns.<br />
<br />
<br />
<br />
[[File:binarypatternrecog.png | 650px|thumb|center|Figure 1: Pattern Memorization experiment - Input Structure and Architecture]]<br />
<br />
<br />
<br />
'''Steps in the experiment:''' <br />
<br />
1) The network is a set of five binary patterns in succession as shown in the figure 1. Each of these patterns has 1,000 elements, for which each element is binary-valued (1 or -1). Here, dark red corresponds to the value 1, and dark blue corresponds to the value -1. <br />
<br />
2) The few shot learning paradigm is followed, where each pattern is shown for 10-time steps, with 3-time steps of zero input between the presentations and the whole sequence of patterns is presented 3 times in random order. <br />
<br />
3) One of the presented patterns is chosen in random order and degraded by setting half of its bits to 0. <br />
<br />
4) This degraded pattern is then fed to the network. The network has to reproduce the correct full pattern in its output using its memory that it developed during training. <br />
<br />
<br />
'''The architecture of the network is described as follows:''' <br />
<br />
1) It is a fully connected RNN with one neuron per pattern element, plus one fixed-output neuron (bias). There are a total of 1,001 neurons. <br />
<br />
2) Value of each neuron is clamped to the value of the corresponding element in the pattern if the value is not 0. If the value is 0, the corresponding neurons do not receive pattern input and must use what it gets from lateral connections and reconstruct the correct, expected output values. <br />
<br />
3) Outputs are read from the activation of the neurons. <br />
<br />
4) The performance evaluation is done by computing the loss between the final network output and the correct expected pattern. <br />
<br />
5) The gradient of the error over the <math display = "inline">w_{i,j}</math> and the <math display = "inline">\alpha_{i,j}</math> coefficients is computed by backpropagation and optimized through Adam solver with learning rate 0.001. <br />
<br />
6) The simple decaying Hebbian formula in Equation 2 is used to update the Hebbian traces. Each network has 2 trainable parameters <math display = "inline">w</math> and <math display = "inline">\alpha</math> for each connection, thus there are a total 1,001 <math display = "inline">\times</math> 1,001 <math display = "inline">\times</math> 2 = 2,004,002 trainable parameters. <br />
<br />
[[File:exp1results.png | 650px|thumb|center|Figure 2:Experiment 1 - Pattern Memorization Results]]<br />
<br />
<br />
The results are shown in the figure 2 where 10 runs are considered. The error becomes quite low after about 200 episodes of training. <br />
<br />
[[File:exp1nonplasticresults.png| 650px|thumb|center|Figure 3: Pattern Memorization results with non plastic networks]]<br />
<br />
<br />
<br />
'''Comparison with Non-Plastic Networks:''' <br />
<br />
1) Non-plastic networks can solve this task but require additional neurons to solve this task in principle. In practice, the authors say that the task is not solved using Non-plastic RNN or LSTM. <br />
<br />
2) The figure 3 shows the results using non-plastic networks. The best results required the addition of 2000 extra neurons. <br />
<br />
3) For non-plastic RNN, the error flattens around 0.13 which is quite high. Using LSTMs, the task can be solved albeit imperfectly and also the error rate reduces drastically t0 around 0.001. <br />
<br />
4) The plastic network solves the task very quickly with the mean error going below 0.01 within 2000 episodes which are mentioned to be 250 times faster than the LSTM.<br />
<br />
= Experiment 2 - Memorizing network images=<br />
<br />
This task is an image reconstruction task that where a network is trained on a set of natural images which it looks to memorize. The natural images with graded pixel values contain more information per element as compared to the last experiment. So this experiment is inherently more complex than the previous ones. Then one image is chosen at random and half the image is displayed to the agent. The task is to complete the image. The paper shows that this method effectively solves this task which other state-of-the-art network architectures fail to solve. <br />
<br />
The experiment is as follows: <br />
<br />
1) Images are from the CIFAR-10 database where there are a total of 60000 images each of size 32 <math display = "inline">\times</math> 32. <br />
<br />
2) The architecture has 1025 neurons in total with a total of 2 <math display = "inline">\times</math> 1025 <math display = "inline">\times</math> 1025 = 2101250 parameters. <br />
<br />
3) Each episode has 3 pictures, shown 3 times for 20-time steps each time, with 3-time steps of zero input between the presentations. <br />
<br />
4) The images are degraded by zeroing out one full contiguous half of the image to prevent a trivial solution of simply reconstructing the missing pixel as the average of its neighbors.<br />
<br />
[[File:exp2results.png| 650px|thumb|center|Figure 4: Natural Image memorization results]]<br />
<br />
<br />
<br />
The results are shown in figure 4. The final output of the network is shown in the last column which is the reconstructed image. The results show that the model has learned to perform this task. <br />
<br />
[[File:exp2weights.png| 650px|thumb|center|Figure 5: Final matrices and plasticity coefficients]]<br />
<br />
The final weight matrix and plasticity coefficients matrix are shown in the figure 5. The plasticity matrix shows a structure related to the high correlation of neighboring pixels and half-field zeroing in test images. <br />
<br />
The full plastic network is compared against a similar architecture with shared plasticity coefficients, where all connections share the same <math display = "inline">\alpha</math> value. So, the single parameter is shared across all connections is trained. <br />
<br />
[[File:independentvsshared.png| 650px|thumb|center|Figure 6: Comparing independent and shared <math display = "inline">\alpha</math> value runs]]<br />
<br />
The figure 6 shows the result of comparison where the independent plasticity coefficient for each connection has better performances. Thus the structure observed in the weight matrices of the results is actually useful.<br />
<br />
<br />
= Experiment 3 - Omniglot task =<br />
<br />
This task involves handwritten symbol recognition. It is a standard task for one-shot and few-shot learning. <br />
<br />
===Experimental Setup: ===<br />
<br />
1) The Omniglot data set is a collection of handwritten characters from various writing systems, including 20 instances each of 1,623 different handwritten characters, written by different subjects.<br />
<br />
[[File:Omniglot Dataset.JPG|400px|center]]<br />
<br />
2) In each episode, N character classes are randomly selected and K instances from each class are sampled. <br />
<br />
3) These instances, together with the class label (from 1 to N), are shown to the model. <br />
<br />
4) Then, a new, unlabeled instance is sampled from one of the N classes and shown to the model.<br />
<br />
5) Model performance is defined as the model’s accuracy in classifying this unlabeled example.<br />
<br />
===Architecture: ===<br />
<br />
1) Model architecture has 4 convolutional layers with 3 <math display = "inline">\times</math> 3 receptive fields and 64 channels. <br />
<br />
2) All convolutions have a stride of 2 to reduce the dimensionality between layers. <br />
<br />
3) The output is a single vector of 64 features, which feeds into an N-way softmax. <br />
<br />
4) The label of the current character is also concurrently fed as a one-hot encoding to this softmax layer, to serve as a guide for the correct output when a label is present.<br />
<br />
===Plasticity in the architecture: ===<br />
<br />
1) Plasticity is applied to the weights from the final layer to the softmax layer, leaving the rest of the convolutional embedding non- plastic. <br />
<br />
2) The expectation is that the convolutional architecture will learn an adequate discriminant between arbitrary handwritten characters and the plastic weights learns to memorize associations between observed patterns and outputs. <br />
<br />
===Data Preparation: ===<br />
<br />
1) The dataset is augmented with rotations by multiples of <math display = "inline">90</math> degrees. <br />
<br />
2) It is divided into 1,523 classes for training and 100 classes (together with their augmentations) for testing. <br />
<br />
3) The networks are trained with an Adam optimizer with a learning rate 3 <math display = "inline">\times 10^{-5}</math>, multiplied by 2/3 every 1M episodes over 5,000,000 episodes. <br />
<br />
4) To evaluate final model performance, 10 models are trained with different random seeds and each of those is tested on 100 episodes using previously unseen test classes.<br />
<br />
===Results: ===<br />
<br />
1) The overall accuracy (i.e. the proportion of episodes with correct classification, aggregated over all test episodes of all runs) is 98.3%, with a 95% confidence interval of 0.80%.<br />
<br />
2) The median accuracy across the 10 runs was 98.5%, indicating consistency in learning.<br />
<br />
{| class="wikitable"<br />
|-<br />
! Memory Networks<br />
! Matching Networks<br />
! ProtoNets<br />
! Memory Module<br />
! MAML<br />
! SNAIL<br />
! DP(This paper)<br />
|-<br />
| 82.8%<br />
| 98.1%<br />
| 97.4%<br />
| 98.4%<br />
| 98.7% <math display = "inline">\pm</math> 0.4<br />
| 99.07% <math display = "inline">\pm</math> 0.16<br />
| 98.03% <math display = "inline">\pm</math> 0.80<br />
|}<br />
<br />
<br />
<br />
3) The above table shows the comparative performance across other non-plastic approaches. The results of the plastic approach are largely similar to those reported for the computationally intensive MAML method and the classification-specialized Matching Networks method. <br />
<br />
4) The performances are slightly below those reported for the SNAIL method, which trains a whole additional temporal-convolution network on top of the convolutional architecture thus having many more parameters.<br />
<br />
5) The conclusion is that a few plastic connections to the output of the network allow for competitive one-shot learning over arbitrary man-made visual symbols.<br />
<br />
= Experiment 4 - Reinforcement learning Maze navigation task =<br />
<br />
This is a maze exploration task where the goal is to teach an agent to reach a goal. The plastic networks are shown to outperform non-plastic ones. <br />
<br />
Experimental setup: <br />
<br />
1) The maze is composed of 9 <math display = "inline">\times</math> 9 squares, surrounded by walls, in which every other square (in either direction) is occupied by a wall. <br />
<br />
[[File:exp4maze.png| 650px|thumb|center|Figure 7: Maze Environment]]<br />
<br />
<br />
2) The maze contains 16 wall square arranged in a regular grid as shown in the figure 7. <br />
<br />
3) At each episode, one non-wall square is randomly chosen as the reward location. When the agent hits this location, it receives a large reward (10.0) and is immediately transported to a random location in the maze Also a small negative reward of -0.1 is provided every time the agent tries to walk into a wall).<br />
<br />
4) Each episode lasts 250-time steps, during which the agent must accumulate as much reward as possible. The reward location is fixed within an episode and randomized across episodes. <br />
<br />
5) The reward is invisible to the agent, and thus the agent only knows it has hit the reward location by the activation of the reward input at the next step.<br />
<br />
6) Inputs to the agent consist of a binary vector describing the 3 <math display = "inline">\times</math> 3 neighborhood centered on the agent (each element being set to 1 or 0 if the corresponding square is or is not a wall), together with the reward at the previous time step. <br />
<br />
7) A2C algorithm is used to meta train the network. <br />
<br />
8) The experiments are run under three conditions: full differentiable plasticity, no plasticity at all, and homogeneous plasticity in which all connections share the same (learnable) <math display = "inline">\alpha</math> parameter. <br />
<br />
9) For each condition, 15 runs with different random seeds are performed. <br />
<br />
<br />
Architecture: <br />
<br />
1) It is a simple recurrent network with 200 neurons, with a softmax layer on top of it to select between the 4 possible actions (up, right, left or down).<br />
<br />
<br />
[[File:exp4performance.png| 650px|thumb|center|Figure 8: Performance curve for the maze navigation experiment]]<br />
<br />
<br />
Results: <br />
<br />
1) The results are shown in the figure 8. The plastic network shows considerably better performance as compared to the other networks.<br />
<br />
2) The non-plastic and homogeneous networks get stuck on a sub-optimal policy. <br />
<br />
3) Thus, the conclusion is that, in this domain, individually sculpting the plasticity of each connection is crucial in reaping the benefits of plasticity for this task.<br />
<br />
= Conclusions =<br />
<br />
<br />
The important contributions from this paper are as follows: <br />
<br />
1) The results show that simple plastic models support efficient meta-learning.<br />
<br />
2) Gradient descent itself is shown to be capable of optimizing the plasticity of a meta-learning system. <br />
<br />
3) The meta-learning is shown to vastly outperform alternative options in the considered experiments. <br />
<br />
4) The method achieved state of the art results on a hard Omniglot test set.<br />
<br />
= Open Source Code =<br />
<br />
Code for this paper can be found at: https://github.com/uber-common/differentiable-plasticity<br />
<br />
<br />
= Critiques =<br />
<br />
The paper addresses an important problem of learning to learn ("meta-learning") and provides a novel framework based on gradient descent to achieve this objective. This paper provides a large scope for future work as many widely used architectures like LSTMs could be tried along with a plastic component. It is also easy to see that the application of such approaches in deep reinforcement learning are also plentiful and there is a good possibility of beating the current baselines in many popular test beds like Atari games using plastic networks. This paper opens up possibilities for a whole class of meta-learning algorithms. <br />
<br />
With regards to the drawbacks of the paper, the paper does not mention how plastic networks will behave if the test sets are completely different from the training dataset. Will the performance be the same as non-plastic networks? It is not very clear if this method will be scalable as there are a large number of parameters to be determined even with the simplest of problems. Also, each experimental domain considered in this paper needed significantly different network architectures (for example in the Omniglot domain plasticity was applied only for the final layers). The paper does not mention any reasons for the specific decisions and if such differences will hold good for other similar problems as well. There has been work in transfer learning applied to both supervised learning and reinforcement learning problems. The authors should have ideally compared plastic networks to performances of some algorithms there as these methods transfer existing knowledge to other related problems and also prevent the need to start training from scratch much similar to the methods adopted in this paper.<br />
<br />
In Experiment 2, the reconstruction of CIFAR-10 images, the authors only provide sample reconstructed images. No quantitative assessment of results is done. It is difficult to judge the generalization of their results. Furthermore, from these results, the authors conclude that their model is good at reconstructing previously unseen images. This claim is quite broad given the relatively simple experiment that was conducted. This is also evident from the network they used, which consisted of only 1000 neurons. Compared with the network in experiment 3, which consisted of a deep 4 layer CNN on a relatively simpler task of classification of Omniglot characters. It would have been more useful if the authors expanded on the image reconstruction task rather than displaying the learned plastic/non-plastic weights. For example, the removed pixels of test images could have been made more random, similar to experiment 1.</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Hierarchical_Representations_for_Efficient_Architecture_Search&diff=40720Hierarchical Representations for Efficient Architecture Search2018-11-22T00:43:27Z<p>Gchalato: /* Introduction */</p>
<hr />
<div>Summary of the paper: [https://arxiv.org/abs/1711.00436 ''Hierarchical Representations for Efficient Architecture Search'']<br />
<br />
= Introduction =<br />
<br />
Deep Neural Networks (DNNs) have shown remarkable performance in several areas such as computer vision, natural language processing, among others; however, improvements over previous benchmarks have required extensive research and experimentation by domain experts. In DNNs, the composition of linear and nonlinear functions produce internal representations of data which are in most cases better than handcrafted ones; consequently, researchers using Deep Learning techniques have lately shifted their focus from working on input features to designing optimal DNN architectures. However, the quest for finding an optimal DNN architecture by combining layers and modules requires frequent trial and error experiments, a task that resembles the previous work on looking for handcrafted optimal features. As researchers aim to solve more difficult challenges the complexity of the resulting DNN is also increasing; therefore, some studies are introducing the use of automated techniques focused on searching for optimal architectures.<br />
<br />
Lately, the use of algorithms for finding optimal DNN architectures has attracted the attention of researchers who have tackled the problem through four main groups of techniques. The first such method employs a supplementary network called a “Hypernet”, which generates ideal network weights given a random architecture. There are two main parts to generating an “optimal” architecture. First, we train the HyperNet. One training cycle consists of generating a random architecture from a sample space of allowed architectures and generating its predicted weights with the HyperNet. Then, the validation score of this proposed network is calculated, and the error is used to backpropogate through the HyperNet. In this manner, the HyperNet can learn to assign robustly optimal initial weights to a given architecture. At “test” time, we generate a random sample of architectures and predict initialized weights for each with our tuned HyperNet. We take the model with the highest validation score and train it as we would a regular architecture. We use this heuristic of “initial validation error” as the relative performance of networks typically stays constant throughout training. That is, if it starts of better, it will very likely end better. The second technique is Monte Carlo Tree Search (MCTS) which repeatedly narrows the search space by focusing on the most promising architectures previously seen. The third group of techniques use evolutionary algorithms where fitness criteria are applied to filter the initial population of DNN candidates, then new individuals are added to the population by selecting the best-performing ones and modifying them with one or several random mutations as in [https://arxiv.org/abs/1703.01041 [Real, 2017]]. The fourth and last group of techniques implement Reinforcement Learning where a policy based controller seeks to optimize the expected accuracy of new architectures based on rewards (accuracy) gained from previous proposals in the architecture space. From these four groups of techniques, Reinforcement Learning has offered the best experimental results; however, the paper we are summarizing implements evolutionary algorithms as its main approach.<br />
<br />
Despite the technique used to look for an optimal architecture, searching in the architecture space usually requires the training and evaluation of many DNN candidates; therefore, it demands huge computational resources and poses a significant limitation for practical applications. Consequently, most techniques narrow the search space with predefined heuristics, either at the beginning or dynamically during the searching process. In the paper we are summarizing, the authors reduce the number of feasible architectures by forcing a hierarchical structure between network components. In other words, each DNN suggested as a candidate is formed by combining basic building blocks to form small modules, then the same basic structures introduced on the building blocks are used to combine and stack networks on the upper levels of the hierarchy. This approach allows the searching algorithm to sample highly complex and modularized networks similar to Inception or ResNet.<br />
<br />
Despite some weaknesses regarding the efficiency of evolutionary algorithms, this study reveals that in fact, these techniques can generate architectures which show competitive performance when a narrowing strategy is imposed over the search space. Accordingly, the main contributions of this paper is a well-defined set of hierarchical representations which acts as the filtering criteria to pick DNN candidates and a novel evolutionary algorithm which produces image classifiers that achieve state of the art performance among similar evolutionary-based techniques.<br />
<br />
=Architecture representations=<br />
<br />
==Flat architecture representation==<br />
All the evaluated network architectures are directed acyclic graphs with only one source and one sink. Each node in the network represents a feature map and consequently, each directed edge represents an operation that takes the feature map in the departing node as input and outputs a feature map on the arriving node. Under the previous assumption, any given architecture in the narrowed search space is formally expressed as a graph assembled by a series of operations (edges) among a defined set of adjacent feature maps (nodes).<br />
<br />
[[File:flatarch.PNG | 650px|thumb|center|Flat architecture representation os neural networks]]<br />
<br />
Multiple primitive operations defined in [https://wiki.math.uwaterloo.ca/statwiki/index.php?title=stat946F18/Hierarchical_Representations_for_Efficient_Architecture_Search#Primitive_operations section 2.3] are used to form small networks defined as ''motifs'' by the authors. To combine the outputs of multiple primitive operations and guarantee a unique output per motif the authors introduce a merge operation which in practice works as a depthwise concatenation that does not require inputs with the same number of channels.<br />
<br />
Accordingly, these motifs can also be combined to form more complex motifs on a higher level in the hierarchy until the network is complex enough to perform competitively in challenging classification tasks.<br />
<br />
==Hierarchical architecture representation==<br />
<br />
The composition of more complex motifs based on simpler motifs at lower levels allows the authors to create a hierarchy-like representation of very complex DNN starting with only a few primitive operations as shown in Figure 1. In other words, an architecture with <math> L </math> levels has only primitive operations at its bottom and only one complex motif at its top. Any motif in between the bottom and top levels can be defined as the composition of motifs in lower levels of the hierarchy.<br />
<br />
[[File:hierarchicalrep.PNG | 700px|thumb|center|Figure 1. Hierarchical architecture representation]]<br />
<br />
In figure 1, the architecture of the full model (its flat structure) is shown in the top right corner. The input (source) is the bottom-most node. The output (sink) is the topmost node. The paper presents an alternative hierarchical view of the model shown on the left-hand side (before the assemble function). This view represents the same model in three layers. The first layer is a set of primitive operations only (bottom row, middle column). In all other layers component motifs (computational graphs) G are described by an adjacency matrix and a set of operations. The set of operations are from the previous layer. An example motif <math> G^{(2)}_{1}</math> in the second layer is shown in the bottom row (left and middle columns). There are three unique motifs in the second layer. These are shown in the middle layer of the top row. Note that the motifs in the previous layer become the operations in the next layer. The higher layer can use these motifs multiple times. Finally, the top level graph, which contains only one motif, <math> G^{(3)}_{1}</math>, is shown in the top row left column. Here, there are 4 nodes with 6 operations defined between them.<br />
<br />
==Primitive operations==<br />
<br />
The six primitive operations used as building blocks for connecting nodes in either flat or hierarchical representations are:<br />
* 1 × 1 convolution of C channels<br />
* 3 × 3 depthwise convolution<br />
* 3 × 3 separable convolution of C channels<br />
* 3 × 3 max-pooling<br />
* 3 × 3 average-pooling<br />
* Identity mapping<br />
<br />
The authors argue that convolution operations involving larger receptive fields can be obtained by the composition of lower-level motifs with smaller receptive fields. Accordingly, convolution operations considering a large number of channels can be generated by the depthwise concatenation of lower-level motifs. Batch normalization and ''ReLU'' activation function are applied after each convolution in the network. There is a seventh operation called null and is used in the adjacency matrix <math> G </math> to state explicitly that there are no operations between two nodes.<br />
<br />
<br />
Side note:<br />
<br />
Some explanations for different types for convolution:<br />
<br />
* Spatial convolution: Convolutions performed in spatial dimensions - width and height.<br />
* Depthwise convolution: Spatial convolution performed independently over each channel of an input.<br />
* 1x1 convolution: Convolution with the kernel of size 1x1<br />
<br />
[[File:convolutions.png | 350px|thumb|center]]<br />
<br />
=Evolutionary architecture search=<br />
<br />
Before moving forward we introduce the concept of genotypes in the context of the article. In this article, a genotype is a particular neural network architecture defined according to the components described in [https://wiki.math.uwaterloo.ca/statwiki/index.php?title=stat946F18/Hierarchical_Representations_for_Efficient_Architecture_Search#Architecture_representations section 2]. In order to make the NN architectures ''evolve'' the authors implemented a three stages process that includes establishing the permitted mutations, creating an initial population and make them compete in a tournament where only the best candidates will survive.<br />
<br />
==Mutation==<br />
<br />
One mutation over a specific architecture is a sequence of five changes in the following order:<br />
<br />
* Sample a level in the hierarchy, different than the basic level.<br />
* Sample a motif in that level.<br />
* Sample a successor node <math>(i)</math> in the motif.<br />
* Sample a predecessor node <math>(j)</math> in the motif.<br />
* Replace the current operation between nodes <math>i</math> and <math>j</math> from one of the available operations.<br />
<br />
The original operation between the nodes <math>i</math> and <math>j</math> in the graph is defined as <math> [G_{m}^{\left ( l \right )}] _{ij} = k </math>. Therefore, a mutation between the same pair of nodes is defined as <math> [G_{m}^{\left ( l \right )}] _{ij} = {k}' </math>.<br />
<br />
The allowed mutations include:<br />
# Change the basic primitive between the predecessor and successor nodes (ie. alter an existing edge): if <math>o_k^{(l-1)} \neq none</math> and <math>o_{k'}^{(l-1)} \neq none</math> and <math>o_{k'}^{(l-1)} \neq >o_k^{(l-1)}</math><br />
# Add a connection between two previously unconnected nodes. The connection between the node can have any of the six possible primitives: if <math>o_k^{(l-1)}=none</math> and <math>o_{k'}^{(l-1)} \neq none</math><br />
# Remove a connection between existing nodes: if <math>o_k^{(l-1)} \neq none</math> and <math>o_{k'}^{(l-1)} = none</math><br />
<br />
==Initialization==<br />
<br />
An initial population is required to start the evolutionary algorithm; therefore, the authors introduced a trivial genotype (candidate solution, the hierarchical architecture of the model) composed only of identity mapping operations. Then a large number of random mutations was run over the ''trivial genotype'' to simulate a diversification process. The authors argue that this diversification process generates a representative population in the search space and at the same time prevents the use of any handcrafted NN structures. Surprisingly, some of these random architectures show a performance comparable to the performance achieved by the architectures found later during the evolutionary search algorithm.<br />
<br />
==Search algorithms==<br />
<br />
Tournament selection and random search are the two search algorithms used by the authors. <br />
<br />
=== Tournament Selection ===<br />
In one iteration of the tournament selection algorithm, 5% of the entire population is randomly selected, trained, and evaluated against a validation set. Then the best performing genotype is picked to go through the mutation process and put back into the population. No genotype is ever removed from the population, but the selection criteria guarantee that only the best performing models will be selected to ''evolve'' through the mutation process.<br />
<br />
We define the pseudocode for tournament selection as follows:<br />
<br />
1. Choose k (the tournament size) individuals from the population at random<br />
<br />
2. Choose the best individual from the tournament with probability p<br />
<br />
3. Choose the second best individual with probability p*(1-p)<br />
<br />
4. Choose the third best individual with probability p*((1-p)^2)<br />
<br />
5. Continue until number of selected individuals equal the number we desire.<br />
<br />
Tournament selection is often chosen over alternative genetic algorithms due to the following benefits: it is efficient to code, works on parallel architectures and allows the selection pressure to be easily adjusted.<br />
<br />
=== Random Search ===<br />
In the random search algorithm every genotype from the initial population is trained and evaluated, then the best performing model is selected. In contrast to the tournament selection algorithm, the random search algorithm is much simpler and the training and evaluation process for every genotype can be run in parallel to reduce search time.<br />
<br />
==Implementation==<br />
<br />
To implement the tournament selection algorithm two auxiliary algorithms are introduced. The first is called the controller and directs the evolution process over the population, in other words, the controller repeatedly picks 5% of genotypes from the current population, send them to the tournament and then apply a random mutation over the best performing genotype from each group. <br />
<br />
[[File:asyncevoalgorithm1.PNG | 700px|thumb|center|Controller]]<br />
<br />
The second auxiliary algorithm is called the worker and is in charge of training and evaluating each genotype, a task that must be completed each time a new genotype is created and added to the population either by an initialization step or by an evolutionary step.<br />
<br />
[[File:asyncevoalgorithm2.PNG | 700px|thumb|center|Worker]]<br />
<br />
Both auxiliary algorithms work together asynchronously and communicate each other through a shared tabular memory file where genotypes and their corresponding fitness are recorded.<br />
<br />
=Experiments and results=<br />
<br />
==Experimental setup==<br />
<br />
Instead of a looking for a complete NN model, the search framework introduced in [https://wiki.math.uwaterloo.ca/statwiki/index.php?title=stat946F18/Hierarchical_Representations_for_Efficient_Architecture_Search#Architecture_representations section 2] is applied to look for the best performing architectures of a small neural network module called the convolutional cell. Using small modules as building blocks to form a larger and more complex model is an approach proved to be successful in previous cases such as the Inception architecture. Additionally, this approach allowed the authors to evaluate cell candidates efficiently and scale to larger and more complex models faster.<br />
<br />
In total three models were implemented as hosts for the experimental cells, the first two use the CIFAR-10 dataset and the third uses the ImageNet dataset. The search framework is implemented only in the first host model to look for the best performing cells ([https://wiki.math.uwaterloo.ca/statwiki/index.php?title=stat946F18/Hierarchical_Representations_for_Efficient_Architecture_Search#Architecture_search_on_CIFAR-10 section 4.2]), once found, these cells were inserted into the second and third host models to evaluate overall performance on the respective datasets ([https://wiki.math.uwaterloo.ca/statwiki/index.php?title=stat946F18/Hierarchical_Representations_for_Efficient_Architecture_Search#Architecture_evaluation_on_CIFAR-10_and_ImageNet section 4.3]).<br />
<br />
The terms training time step, initialization time step, and evolutionary time step will be used to describe some parts of the experiments. Be aware that these three terms have different meanings; however, each term will be properly defined when introduced.<br />
<br />
==Architecture search on CIFAR-10==<br />
<br />
The overall goal in this stage is to find the best performing cells. The search framework is run using the small CIFAR-10 depicted in Figure 2 as host model for the cells; therefore, during the searching process, only the cells change while the rest of the host model’s structure remains the same. In the context of the evolutionary search algorithm, a cell is also called a candidate or a genotype. Additionally, on every time step during the search process, the three cells in the model will share the same structure and consequently every time a new candidate architecture is evaluated the three cells will simultaneously adopt the new candidate’s architecture.<br />
<br />
[[File:smallcifar10.PNG | 350px|thumb|center|Figure 2. Small CIFAR-10 model]]<br />
<br />
To begin the architecture searching process an initial population of genotypes is required. Random mutations are applied over a trivial genotype to generate a candidate and grow the seminal population. This is called an initialization step and is repeated 200 times to produce an equivalent number of candidates. Creating these 200 candidates with random structures is equivalent to running a random search over a constrained architecture space. <br />
<br />
Then, the evolutionary search algorithm takes over and runs from timestep 201 up to time step 7000, these are called evolutionary timesteps. On each evolutionary time step, a group of genotypes equivalent to 5% of the current population is selected randomly and sent to the tournament for fitness computation. To perform a fitness evaluation each candidate cell is inserted into the three predefined positions within the small CIFAR-10 host model. Then for each candidate cell, the host model is trained with stochastic gradient descent during 5000 training steps and decreasing learning rate. Due to a small standard deviation of up to 0.2% found when evaluating the exact same model, the overall fitness is obtained as the average of four training-evaluation runs. Finally, a random mutation is applied over a copy of the best cell within the group to create a new genotype that is added to the current population.<br />
<br />
The fitness of each evaluated genotype is recorded in the shared tabular memory file to avoid recalculation in case the same genotype is selected again in a future evolutionary time step.<br />
<br />
The search framework is run for 7000-time steps (200 initialization time steps and the rest are evolutionary time steps) for each one of three different types of cell architecture, namely hierarchical representation, flat representation and flat representation with constrained parameters. <br />
<br />
* A cell that follows a hierarchical representation has NN connections at three different levels; at the bottom level it has only primitive operations, at the second level it contains motifs with four-nodes and at the third level it has only one motif with five-nodes.<br />
<br />
* A cell that follows a flat representation has 11 nodes with only primitive operations between them. These cells look similar to level 2 motifs but instead of having four nodes they have 11 and therefore many more pairs of nodes and operations.<br />
<br />
* For a cell that follows a flat representation with constrained parameters the total number of parameters used by its operations cannot be superior to the total number of parameters used by the cells that follow a hierarchical representation.<br />
<br />
Figure 3 shows the current fitness achieved by the best performing cell from each one of the three types of cells when plugged in the small CIFAR-10 model. Even though the fitness grows rapidly after the first 200 (initialization) time steps, it tends to plateau between 89% to 90%. Overall, cells that follow a flat representation without restriction in the number of parameters tend to perform better than those following a hierarchical structure. It could be due to the fact that the flat representation allows more flexibility when adding connections between nodes, especially between distant ones. Unfortunately, the authors do not describe the architecture of the best performing flat cell.<br />
<br />
[[File:currentfitness.PNG | 300px|thumb|center|Figure 3. Current fitness]]<br />
<br />
Figure 4 presents the maximum fitness reached by any cell seen by the search framework between each one of the three types of cells, the fitness at time step 200 is, therefore, equivalent to the best model obtained by a random search over 200 architectures from each type of cell.<br />
<br />
[[File:maxfitness.PNG | 300px|thumb|center|Figure 4. Maximum fitness]]<br />
<br />
The total number of parameters used by each genotype at any given time step is shown in Figure 5. It suggests that flat representations tend to add more connections over time and most likely those connections correspond to convolutional operations which in turn require more parameters than other primitive operations.<br />
<br />
[[File:numparameters.PNG | 300px|thumb|center|Figure 5. Number of parameters]]<br />
<br />
To run each time step (either initialization or evolutionary) in the search framework, it takes one hour for a GPU to perform four training and evaluation rounds for every single candidate. Therefore, the authors used 200 GPUs simultaneously to complete 7000-time steps in 35 hours. Considering the three types of cell (hierarchical, flat, and parameter-constrained flat), approximately 20000 GPU-hours could be required to replicate the experiment.<br />
<br />
==Architecture evaluation on CIFAR-10 and ImageNet==<br />
<br />
Once the evolutionary search finds the best-fitted cells those are plug into the two larger host models to evaluate their performance in those more complex architectures. The first large model (Figure 6) is targeted to image classification on the CIFAR-10 dataset and the second model (Figure 7) is focused on image classification on the ImageNet dataset. Although all the parameters in these two larger host models are trained from scratch including those within the cells, no changes in the cell’s architectures will happen since their structure was found to be optimal during the evolutionary search.<br />
<br />
The large CIFAR-10 model is trained with stochastic gradient descent during 80K training steps and decreasing learning rate. To account for the non-negligible standard deviation found when evaluating the exact same model, the percentage of error is determined as the average of five training-evaluation runs.<br />
<br />
[[File:largecifar10.PNG | 500px|thumb|center|Figure 6. Large CIFAR-10 model]]<br />
<br />
The ImageNet model is trained with stochastic gradient descent during 200K training steps and decreasing learning rate. For this model, neither standard deviation nor multiple training-evaluation runs were reported.<br />
<br />
[[File:imagenetmodel.PNG | 600px|thumb|center|Figure 7. ImageNet model]]<br />
<br />
In [https://wiki.math.uwaterloo.ca/statwiki/index.php?title=stat946F18/Hierarchical_Representations_for_Efficient_Architecture_Search#Architecture_search_on_CIFAR-10 section 4.2] three types of cells were described: hierarchical, flat, and parameter-constrained flat. For the hierarchical type of cells, the percentage of error in both large models is reported in Table 1 for four different cases: a cell with random architecture, the best-fitted cell from 200 random architectures, the best-fitted cell from 7000 random architectures, and the best-fitted cell after 7000 evolutionary steps. On the other hand, for the flat and parameter-constrained flat types of architecture, only some of the mentioned four cases are reported in Table 1.<br />
<br />
[[File:comparisoncells.PNG | 750px|thumb|center|Table 1. Comparison between types of cells and searching method]]<br />
<br />
According to the results in Table 1, for both large host models, the hierarchical cell found by the evolutionary search algorithm achieved the lowest errors with 3.75% in CIFAR-10, 20.3% top-1 error and 5.2% top-5 error in ImageNet. The errors reported in both datasets are calculated by using the trained large models on test sets of images never seen before during any of the previous stages. Even though the cell that follows a hierarchical representation achieved the lowest error, the ones showing the lowest standard deviations are those following a flat representation.<br />
<br />
The performance achieved by the large CIFAR-10 host model using the best cell is then compared against other classifiers in Table 2. As an additional improvement, the authors increased the number of channels in its first convolutional layer from 64 to 128. It is worth to note that this first convolutional layer is not part of the cell obtained during the evolutionary search process, instead, it is part of the original host model. The results are grouped into three categories depending on how the classifiers involved in the comparison were created, from top to bottom: handcrafted, reinforcement learning, and evolutionary algorithms.<br />
<br />
[[File:comparisonlargecifar10.PNG | 500px|thumb|center|Table 2. Comparison against other classifiers on CIFAR-10]]<br />
<br />
The classification error achieved by the ImageNet host model when using the best cell is also compared against some high performing image classifiers in the literature and the results are presented in Table 3. Although the classification error scored by the architecture introduced in this paper is not significantly lower than those obtained by state of the art classifiers, it shows outstanding results considering that it is not a hand engineered structure.<br />
<br />
[[File:comparisonimagenet.PNG | 500px|thumb|center|Table 3. Comparison against other classifiers on ImageNet]]<br />
<br />
A visualisation of the evolved hierarchical cell is shown below. The detailed visualisations of each motif can be seen in Appendix A of the paper. It can be noted that motif 4 directly links the input and output, and itself contains (among other operations) an identity mapping from input to output. Many other such 'skip connections' can be seen.<br />
<br />
[[File:WF_SecCont_03_hier_vis.png]]<br />
<br />
=Conclusion=<br />
<br />
A new evolutionary framework is introduced for searching neural network architectures over searching spaces defined by flat and hierarchical representations of a convolutional cell, which uses smaller operations instead of the larger ones as the building blocks. Experiments show that the proposed framework achieves competitive results against state of the art classifiers on the CIFAR-10 and ImageNet datasets.<br />
<br />
=Critique=<br />
<br />
While the method introduced in this paper achieves a lower error in comparison to other evolutionary methods, it is not significantly better than those obtained by handcrafted design or reinforcement learning. A more in-depth analysis considering the number of parameters and required computational resources would be necessary to accurately compare the listed methods.<br />
<br />
In [https://wiki.math.uwaterloo.ca/statwiki/index.php?title=Hierarchical_Representations_for_Efficient_Architecture_Search#Architecture_evaluation_on_CIFAR-10_and_ImageNet section 4.3] it is not clear why the results for the four different cases that are reported for the hierarchical cells in Table 1 are not reported for the ones following a flat representation, considering that the flat cells showed a better performance during the evolutionary search. Recall that the four cases are: a cell with random architecture, the best-fitted cell from 200 random architectures, the best-fitted cell from 7000 random architectures, and the best-fitted cell after 7000 evolutionary steps.<br />
<br />
It seems contradictory that the flat type of cells who clearly performed better than the hierarchical ones during the architecture search ([https://wiki.math.uwaterloo.ca/statwiki/index.php?title=Hierarchical_Representations_for_Efficient_Architecture_Search#Architecture_search_on_CIFAR-10 section 4.2]) are not the ones scoring the lowest error when evaluated on the two large host models ([https://wiki.math.uwaterloo.ca/statwiki/index.php?title=Hierarchical_Representations_for_Efficient_Architecture_Search#Architecture_evaluation_on_CIFAR-10_and_ImageNet section 4.3]).<br />
<br />
= References =<br />
<br />
# Hanxiao Liu, Karen Simonyan, Oriol Vinyals, Chrisantha Fernando, Koray Kavukcuoglu, https://arxiv.org/abs/1711.00436.</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Pixels_to_Graphs_by_Associative_Embedding&diff=37355Pixels to Graphs by Associative Embedding2018-10-30T14:18:10Z<p>Gchalato: </p>
<hr />
<div>== Introduction == <br />
The paper presents a novel approach to generating a scene graph. A scene graph, as it relates to an image, is a graph with a vertex that represents each object identified in the image and an edge that represents relationships between the objects. <br />
<br />
An example of a scene graph:<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Scene Graph.PNG]]</div><br />
<br />
Current state-of-the-art techniques break down the construction of scene graphs by first identifying objects, and then predicting the edges for any given pair of identified objects. By using this technique, reasoning over<br />
the full graph would be limited. On the other hand, this paper introduces an architecture that defines the entire graph directly from the image, enabling the network to reason across the entirety of the image to understand relationships, as opposed to only predicting relationships using object labels. <br />
<br />
A key concern, given that the new architecture produces both vertices (objects) and edges (relationships), is connecting the two. Specifically, the output of the network is some set of relationships E, and some set of Vertices V. The network needs to also output the “source” and “destination” of each relationship, so that the final graph can be formed. In the image above, for example, the network would also need to tell us that “holding” comes from “person” and goes to “Frisbee”. To do this, the paper uses associative embeddings. Specifically, the network outputs a particular “embedding vector” for each vertex, as well as a “source embedding” and “destination embedding” for each relationship. A final post-processing step finds the vertex embedding closest to each of the source/destination embeddings of each relationship and in this way assigns the edges to pairs of vertices.<br />
<br />
== Previous Works == <br />
<br />
In the field of relationship detection, the following are the existing state of the art advances:<br />
<br />
1) Framing the task of identifying objects using localization from referential expressions, detection of human-object interactions, or the more general tasks of visual relationship detection (VRD) and scene graph generation. <br />
<br />
2) Visual relationship detection methods like message passing RNNs and predicting over triplets of bounding boxes. <br />
<br />
In the field of associative embedding, the following are some interesting applications: <br />
<br />
1) Vector embeddings to group together body joints for multiperson pose estimation. <br />
<br />
2) Vector embeddings to detect body joints of the various people in an image.<br />
<br />
<br />
<br />
<br />
== The Architecture: == <br />
: '''1. Detecting Graph Elements'''<br />
<br />
Given an image of dimensions h x w, a stacked hourglass (Appendix 2) is used to generate a h x w x f representation of the image. It should be noted that the dimension of the output (which is non-trainable),needs to fulfill certain criteria. Specifically, we need to have a resolution large enough to minimize the number of pixels with multiple detections while also being small enough to ensure that each 1 x 1 x f vector still contains the information needed for subsequent inference.<br />
<br />
A 1x1 convolution and sigmoid activation is performed on this result to generate a heat map (one for objects and one for relationships, using separately determined convolutions). The value at a given pixel can be interpreted as the likelihood of detection at that particular pixel in the original image. <br />
<br />
In order to claim that there is an element at some pixel, we need to have some likelihood threshold. Then, if a given pixel in the map has a value >= the threshold, we claim that there is an element at that pixel. This threshold is calculated by using binary cross-entropy loss on the final values in the heatmap. Values with likelihoods greater than p-hat will be considered element detections. <br />
<br />
Finally, for each element that we detected, we extract the 1 x 1 x f feature vector. This is then used as an input to a set of Feed Forward Neural Networks (FFNNs), where we have a separate network for each characteristic of interest, and for each network, there's one hidden layer with f nodes. The object class and relationship (edges) could be supervised by softmax loss. Furthermore, in order to predict the bounding box of the object, we can use the approach in Faster-RCNN[3]. The following image summarizes the process.<br />
<br />
<br />
[[File:Extraction Process.PNG|center]]<br />
<br />
:'''2. Connecting Elements with Associative Embeddings'''<br />
As explained earlier, to construct the scene graph, we need to know the source and destination of each edge. This is done through associative embeddings. <br />
<br />
First, let us define an embedding hi ϵ Rd produced for some vector i, and let us assume that we have n object detections in a particular image. Now, define hik, for k = 1 to Ki (where Ki is the number of edges in the graph with a vertex at vertex i) as the embedding associated with an edge that touches vertex i. We define two loss functions on these sets.<br />
<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Loss 1.PNG]]</div><br />
<br />
The goal of Lpull is to minimize the squared differences between the embedding of a given vertex and the embedding of an edge that references said vertex.<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Loss 2.PNG]]</div><br />
<br />
On the other hand, minimizing Lpush implies assigning embeddings to vertices that are as far apart as possible. The further apart they are, the lower the output of max becomes until eventually, it reaches 0. Here, m is just a constant. In the paper, the values used were m = 8 and d = 8 (that is, 8D embeddings). Combining these two loss functions (and weighing them equally), accomplishes the task of predicting embeddings such that vertices are differentiated, but the embedding of a vertex is most similar to the vertex it references.<br />
<br />
:'''3. Support for Overlapping Detections'''<br />
An obvious concern is how the network would operate if there was more than one detection (be it object or relationship), in a given pixel. For example, detection of “shirt” and “person” may be centered at the exact same pixel. To account for this, the architecture is modified to allow for “slots” at each pixel. Specifically, so detections of objects are allowed at a particular pixel, while sr relationship detections are allowed at a given pixel. <br />
<br />
In order to allow for this, some changes are required after the feature extraction step. Specifically, we now use the 1x1xf vector as the input for so (or sr) different sets of 4 FFNNs, where the output (of the first three) is as shown in figure 2, and with the final FFNN outputting the probability of a detection existing in that particular slot, at that particular pixel. This new network is trained exclusively on whether or not a detection has been made in that slot, and, in prediction, is used to determine the number of slots to output at a given pixel. It is critical to note that this each of these so (or sr) sets of FFNNs share absolutely no weights. And each is trained for detection in its assigned slot.<br />
<br />
It is important to note that this implies a change in the training procedure. We now have so (or sr) different predictions (be it class, or class + bounding box), that we need to match with our set of ground truth detections at a given pixel. Without this step, we would not be able to assign a value to the error for that sample. To do this, we match a one-hot encoded vector of the ground-truth class and bounding box anchor (the reference vector), and then match them with the so (or sr) outputs provided at a given pixel. The Hungarian method is used to ensure maximum matching between the outputs and the reference method while ensuring we do not assign the same detection to multiple slots.<br />
<br />
==Results==<br />
A quick note on notation: R@50 indicates what percentage of ground-truth subject-predicate-object tuples appeared in a proposal of 50 such tuples. Since R@100 offers more possibilities, it will necessarily be higher. The 6.7, for example, indicates that 6.7% of the ground truth tuples appeared in the proposals of the network. <br />
<br />
The authors tested the network against two other architectures designed to develop a semantic understanding of images. For this, they used the Visual Genome dataset, with so = 3 and sr = 6. Overall, the new architecture vastly outperformed past models. The results were as follows:<br />
<br />
The table can be interpreted as follows:<br />
<br />
[[File:Results Table.PNG|center|600px]]<br />
<br />
::'''SGGen (no RPN)''': Given a particular image, without the use of Region Proposal networks, the accuracy of the proposed scene graph. No class predictions are provided.<br />
::'''SGGen (with RPN)''': Same as above, except the output of the Region Proposal Network, is used to enhance the input of a given image. No class predictions are provided.<br />
::'''SGCIs''': Ground-truth object bounding boxes are provided. The network is asked to classify them and determine relationships.<br />
::'''PredCIs''': As above, except the classes are also provided. The only goal is to predict relationships.<br />
<br />
Further analysis into the accuracy, when looking at predicates individually, shows that the architecture is very sensitive to over-represented relationship predicates.<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Results - Part 2.PNG]]</div><br />
<br />
As shown in Figure 5, for many ground-truth predicates (those that do not appear often in the ground truth), the network does poorly. Even when allowed to propose 100 tuples, the network does not offer the predicate. Figure 4 simply observes the fact that certain sets of relationship predicates appear predominantly in a subset of slots. No general explanation has been offered for this behavior.<br />
<br />
== Conclusion ==<br />
In conclusion, the paper offers a novel approach that enables the extraction of image semantics while perpetually reasoning over the entire context of the image. Associative embeddings are used to connect object and predicate relationships, and parallel “slots” allow for multiple detections in one pixel. While this approach offers noticeable improvements in accuracy, it is clear that work needs to be done to account for the non-uniform distributions of relationships in the dataset.<br />
<br />
<br />
== Critiques ==<br />
<br />
The paper's contributions towards patterning unordered network outputs and using associative embeddings for connecting vertices and edges are commendable. However, it should be noted this paper is only an incremental improvement over existing well-studied architectures like the hour glass architecture. The modifications also seem to be hacky. The authors say that they make a slight modification to the hourglass design and double the number of features and weight all the loses equally. No scientific justification for why this is needed is given. Also the choice of constants to be 3 and 6 for <math display = "inline"> s_o</math> and <math display = "inline"> s_r</math> is not clear, as the authors leave out a fraction of the cases. I am not sure if the changes made are truly a critical advance as the experiments are conducted only on a single dataset and no generalizability arguments are made by the authors. So the methods might just work well only for this dataset and the changes may pertain to only this one. The theoretical analysis done in the paper comes directly from the hourglass literature and cannot be accounted for novelty. <br />
== Appendices ==<br />
<br />
'''Appendix 1: Sample Outputs'''<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Sample Pixel Graph Outputs.PNG]]</div><br />
<br />
'''Appendix 2: Stacked Hourglass Architecture'''<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Hourglass.PNG]]</div><br />
<br />
Although this goes beyond the focus of the paper, I would like to add a brief overview of the stacked hourglass architecture used to generate the heatmap. This architecture is unique in that it allows cyclical top-down, bottom-up inference and recombination of features. While most architectures focus on optimizing the bottom-up portion (reducing dimensionality), the stacked-hourglass gives the network more flexibility in how it generates a representation by allowing it to learn a series of down-sampling / up-sampling steps.<br />
<br />
== References ==<br />
1. Alejandro Newell and Jia Deng, “Pixels to Graphs by Associative Embedding,” in NIPS, 2017<br />
<br />
2. Alejandro Newell, Kaiyu Yang, and Jia Deng. Stacked Hourglass Networks for Human Pose Estimation. ECCV, 2016<br />
<br />
3. Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real-time object detection with region proposal networks. NIPS, pages 91–99, 2015.</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Pixels_to_Graphs_by_Associative_Embedding&diff=37175Pixels to Graphs by Associative Embedding2018-10-24T02:49:49Z<p>Gchalato: /* The Architecture: */</p>
<hr />
<div>== Introduction == <br />
The paper presents a novel approach to proposing a scene graph. A scene graph, as it relates to an image, is a graph with a vertex that represents each object identified in the image and an edge that represents relationships between the objects. <br />
<br />
An example of a scene graph:<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Scene Graph.PNG]]</div><br />
<br />
Current state-of-the-art techniques break down the construction of scene graphs by first identifying objects, and then predicting the edges for any given pair of identified objects. This paper introduces an architecture that defines the entire graph directly from the image, enabling the network to reason across the entirety of the image to understand relationships, as opposed to only predicting relationships using object labels. <br />
<br />
A key concern, given that the new architecture produces both vertices (objects) and edges (relationships) , is connecting the two. Specifically, the output of the network is some set of relationships E, and some set of Vertices V. The network needs to also output the “source” and “destination” of each relationship, so that the final graph can be formed. In the image above, for example, the network would also need to tell us that “holding” comes from “person” and goes to “Frisbee”. To do this, the paper uses associative embeddings. Specifically, the network outputs a particular “embedding vector” for each vertex, as well as a “source embedding” and “destination embedding” for each relationship. A final post-processing step finds the vertex embedding closest to each of the source / destination embeddings of each relationship and in this way assigns the edges to pairs of vertices.<br />
<br />
== The Architecture: == <br />
: '''1. Detecting Graph Elements'''<br />
<br />
Given an image of dimensions h x w, a stacked hourglass (Appendix 2) is used to generate a h x w x f representation of the image. It should be noted that the dimension of the output (which is non-trainable), to consider. Specifically, we need to have a resolution large enough to minimize the number of pixels with multiple detections while also being small enough to ensure that each 1 x 1 x f vector still contains the information needed for subsequent inference.<br />
<br />
A 1x1 convolution and sigmoid activation is performed on this result to generate a heatmap (one for objects and one for relationships, using separately determined convolutions). The value at a given pixel can be interpreted as the likelihood of detection at that particular pixel in the original image. <br />
<br />
In order to claim that there is an element at some pixel, we need to have some likelihood threshold. Then, if a given pixel in the map has a value >= the threshold, we claim that there is an element at that pixel. This threshold is calculated by using binary cross-entropy loss on the final values in the heatmap. Values with likelihoods greater than p-hat will be considered element detections. <br />
<br />
Finally, for each element that we detected, we extract the 1 x 1 x f feature vector. This is then used as an input to a set of Feed Forward Neural Networks (FFNNs), where we have a separate network for each characteristic of interest. The following image summarizes the process.<br />
<br />
<br />
[[File:Extraction Process.PNG|center]]<br />
<br />
:'''2. Connecting Elements with Associative Embeddings'''<br />
As explained earlier, to construct the scene graph, we need to know the source and destination of each edge. This is done through associative embeddings. <br />
<br />
First, let us define an embedding hi ϵ Rd produced for some vector i, and let us assume that we have n object detections in a particular image. Now, define hik, for k = 1 to Ki (where Ki is the number of edges in the graph with a vertex at vertex i) as the embedding associated with an edge that touches vertex i. We define two loss functions on these sets.<br />
<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Loss 1.PNG]]</div><br />
<br />
The goal of Lpull is minimize the squared differences between the embedding of a given vertex and the embedding of an edge that references said vertex.<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Loss 2.PNG]]</div><br />
<br />
On the other hand, minimizing Lpush implies assigning embeddings to vertices that are as far apart as possible. The further apart they are, the lower the output of max becomes, until eventually it reaches 0. Here, m is just a constant. In the paper, the values used were m = 8 and d = 8 (that is, 8D embeddings). Combining these two loss functions (and weighing them equally), accomplishes the task of predicting embeddings such that vertices are differentiated, but the embedding of a vertex is most similar to the vertex it references.<br />
<br />
:'''3. Support for Overlapping Detections'''<br />
An obvious concern is how the network would operate if there was more than one detection (be it object or relationship), in a given pixel. For example, detection of “shirt” and “person” may be centered at the exact same pixel. To account for this, the architecture is modified to allow for “slots” at each pixel. Specifically, so detections of objects are allowed at a particular pixel, while sr relationship detections are allowed at a given pixel. <br />
<br />
In order to allow for this, some changes are required after the feature extraction step. Specifically, we now use the 1x1xf vector as the input for so (or sr) different sets of 4 FFNNs, where the output (of the first three) is as shown in figure 2, and with the final FFNN outputting the probability of a detection existing in that particular slot, at that particular pixel. This new network is trained exclusively on whether or not a detection has been made in that slot, and, in prediction, is used to determine the number of slots to output at a given pixel. It is critical to note that this each of these so (or sr) sets of FFNNs share absolutely no weights. And each is trained for detection in its assigned slot.<br />
<br />
It is important to note that that this implies a change in the training procedure. We now have so (or sr) different predictions (be it class, or class + bounding box), that we need to match with our set of ground truth detections at a given pixel. Without this step, we would not be able to assign a value to the error for that sample. To do this, we match a one-hot encoded vector of the ground-truth class and bounding box anchor (the reference vector), and then match them with the so (or sr) outputs provided at a given pixel. The Hungarian method is used to ensure maximum matching between the outputs and the reference method while ensuring we do not assign the same detection to multiple slots.<br />
<br />
==Results==<br />
A quick note on notation: R@50 indicates what percentage of ground-truth subject-predicate-object tuples appeared in a proposal of 50 such tuples. Since R@100 offers more possibilities, it will necessarily be higher. The 6.7, for example, indicates that 6.7% of the ground truth tuples appeared in the proposals of the network. <br />
<br />
The authors tested the network against two other architectures designed to develop semantic understanding of images. For this, they used the Visual Genome dataset, with so = 3 and sr = 6. Overall, the new architecture vastly outperformed past models. The results were as follows:<br />
<br />
The table can be interpreted as follows:<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Results Table.PNG]]</div><br />
<br />
::'''SGGen (no RPN)''': Given a particular image, without the use of Region Proposal networks, the accuracy of the proposed scene graph. No class predictions are provided.<br />
::'''SGGen (with RPN)''': Same as above, except the output of the Region Proposal Network is used to enhance the input of a given image. No class predictions are provided.<br />
::'''SGCIs''': Ground-truth object bounding boxes are provided. The network is asked to classify them and determine relationships.<br />
::'''PredCIs''': As above, except the classes are also provided. The only goal is to predict relationships.<br />
<br />
Further analysis into the accuracy, when looking at predicates individually, shows that the architecture is very sensitive to over-represented relationship predicates.<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Results - Part 2.PNG]]</div><br />
<br />
As shown in Figure 5, for many ground-truth predicates (those that do not appear often in the ground truth), the network does poorly. Even when allowed to propose 100 tuples, the network does not offer the predicate. Figure 4 simply observes the fact that certain sets of relationship predicates appear predominantly in a subset of slots. No general explanation has been offered for this behaviour.<br />
<br />
== Conclusion ==<br />
In conclusion, the paper offers a novel approach to that enables the extraction of image semantics while perpetually reasoning over the entire context of the image. Associative embeddings are used to connect object and predicate relationships, and parallel “slots” allow for multiple detections in one pixel. While this approach offers noticeable improvements in accuracy, it is clear that work needs to be done to account for the non-uniform distributions of relationships in the dataset.<br />
<br />
== Appendices ==<br />
<br />
'''Appendix 1: Sample Outputs'''<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Sample Pixel Graph Outputs.PNG]]</div><br />
<br />
'''Appendix 2: Stacked Hourglass Architecture'''<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Hourglass.PNG]]</div><br />
<br />
Although this goes beyond the focus of the paper, I would like to add a brief overview of the stacked hourglass architecture used to generate the heatmap. This architecture is unique in that it allows cyclical top-down, bottom-up inference and recombination of features. While most architectures focus on optimizing the bottom-up portion (reducing dimensionality), the stacked-hourglass gives the network more flexibility in how it generates a representation by allowing it to learn a series of down-sampling / up-sampling steps.<br />
<br />
== References ==<br />
1. Alejandro Newell and Jia Deng, “Pixels to Graphs by Associative Embedding,” in NIPS, 2017<br />
<br />
2. Alejandro Newell, Kaiyu Yang, and Jia Deng. Stacked Hourglass Networks for Human Pose Estimation. ECCV, 2016</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Pixels_to_Graphs_by_Associative_Embedding&diff=37173Pixels to Graphs by Associative Embedding2018-10-24T02:46:07Z<p>Gchalato: /* References */</p>
<hr />
<div>== Introduction == <br />
The paper presents a novel approach to proposing a scene graph. A scene graph, as it relates to an image, is a graph with a vertex that represents each object identified in the image and an edge that represents relationships between the objects. <br />
<br />
An example of a scene graph:<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Scene Graph.PNG]]</div><br />
<br />
Current state-of-the-art techniques break down the construction of scene graphs by first identifying objects, and then predicting the edges for any given pair of identified objects. This paper introduces an architecture that defines the entire graph directly from the image, enabling the network to reason across the entirety of the image to understand relationships, as opposed to only predicting relationships using object labels. <br />
<br />
A key concern, given that the new architecture produces both vertices (objects) and edges (relationships) , is connecting the two. Specifically, the output of the network is some set of relationships E, and some set of Vertices V. The network needs to also output the “source” and “destination” of each relationship, so that the final graph can be formed. In the image above, for example, the network would also need to tell us that “holding” comes from “person” and goes to “Frisbee”. To do this, the paper uses associative embeddings. Specifically, the network outputs a particular “embedding vector” for each vertex, as well as a “source embedding” and “destination embedding” for each relationship. A final post-processing step finds the vertex embedding closest to each of the source / destination embeddings of each relationship and in this way assigns the edges to pairs of vertices.<br />
<br />
== The Architecture: == <br />
: '''1. Detecting Graph Elements'''<br />
<br />
Given an image of dimensions h x w, a stacked hourglass (Appendix 2) is used to generate a h x w x f representation of the image. It should be noted that the dimension of the output (which is non-trainable), to consider. Specifically, we need to have a resolution large enough to minimize the number of pixels with multiple detections while also being small enough to ensure that each 1 x 1 x f vector still contains the information needed for subsequent inference.<br />
<br />
A 1x1 convolution and sigmoid activation is performed on this result to generate a heatmap (one for objects and one for relationships, using separately determined convolutions). The value at a given pixel can be interpreted as the likelihood of detection at that particular pixel in the original image. <br />
<br />
In order to claim that there is an element at some pixel, we need to have some likelihood threshold. Then, if a given pixel in the map has a value >= the threshold, we claim that there is an element at that pixel. This threshold is calculated by using binary cross-entropy loss on the final values in the heatmap. Values with likelihoods greater than p-hat will be considered element detections. <br />
<br />
Finally, for each element that we detected, we extract the 1 x 1 x f feature vector. This is then used as an input to a set of FFNN, where we have a separate network for each characteristic of interest. The following image summarizes the process.<br />
<br />
<br />
[[File:Extraction Process.PNG|center]]<br />
<br />
:'''2. Connecting Elements with Associative Embeddings'''<br />
As explained earlier, to construct the scene graph, we need to know the source and destination of each edge. This is done through associative embeddings. <br />
<br />
First, let us define an embedding hi ϵ Rd produced for some vector i, and let us assume that we have n object detections in a particular image. Now, define hik, for k = 1 to Ki (where Ki is the number of edges in the graph with a vertex at vertex i) as the embedding associated with an edge that touches vertex i. We define two loss functions on these sets.<br />
<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Loss 1.PNG]]</div><br />
<br />
The goal of Lpull is minimize the squared differences between the embedding of a given vertex and the embedding of an edge that references said vertex.<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Loss 2.PNG]]</div><br />
<br />
On the other hand, minimizing Lpush implies assigning embeddings to vertices that are as far apart as possible. The further apart they are, the lower the output of max becomes, until eventually it reaches 0. Here, m is just a constant. In the paper, the values used were m = 8 and d = 8 (that is, 8D embeddings). Combining these two loss functions (and weighing them equally), accomplishes the task of predicting embeddings such that vertices are differentiated, but the embedding of a vertex is most similar to the vertex it references.<br />
<br />
:'''3. Support for Overlapping Detections'''<br />
An obvious concern is how the network would operate if there was more than one detection (be it object or relationship), in a given pixel. For example, detection of “shirt” and “person” may be centered at the exact same pixel. To account for this, the architecture is modified to allow for “slots” at each pixel. Specifically, so detections of objects are allowed at a particular pixel, while sr relationship detections are allowed at a given pixel. <br />
<br />
In order to allow for this, some changes are required after the feature extraction step. Specifically, we now use the 1x1xf vector as the input for so (or sr) different sets of 4 FFNNs, where the output is as shown in figure 2, but with the final FFNN outputting the probability of a detection existing in that particular slot, at that particular pixel. This new network is trained exclusively on whether or not a detection has been made in that slot, and, in prediction, is used to determine the number of slots to output at a given pixel. It is critical to note that this each of these so (or sr) sets of FFNNs share absolutely no weights. And each is trained for detection in its assigned slot.<br />
<br />
It is important to note that that this implies a change in the training procedure. We now have so (or sr) different predictions (be it class, or class + bounding box), that we need to match with our set of ground truth detections at a given pixel. Without this step, we would not be able to assign a value to the error for that sample. To do this, we match a one-hot encoded vector of the ground-truth class and bounding box anchor (the reference vector), and then match them with the so (or sr) outputs provided at a given pixel. The Hungarian method is used to ensure maximum matching between the outputs and the reference method while ensuring we do not assign the same detection to multiple slots. <br />
<br />
<br />
==Results==<br />
A quick note on notation: R@50 indicates what percentage of ground-truth subject-predicate-object tuples appeared in a proposal of 50 such tuples. Since R@100 offers more possibilities, it will necessarily be higher. The 6.7, for example, indicates that 6.7% of the ground truth tuples appeared in the proposals of the network. <br />
<br />
The authors tested the network against two other architectures designed to develop semantic understanding of images. For this, they used the Visual Genome dataset, with so = 3 and sr = 6. Overall, the new architecture vastly outperformed past models. The results were as follows:<br />
<br />
The table can be interpreted as follows:<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Results Table.PNG]]</div><br />
<br />
::'''SGGen (no RPN)''': Given a particular image, without the use of Region Proposal networks, the accuracy of the proposed scene graph. No class predictions are provided.<br />
::'''SGGen (with RPN)''': Same as above, except the output of the Region Proposal Network is used to enhance the input of a given image. No class predictions are provided.<br />
::'''SGCIs''': Ground-truth object bounding boxes are provided. The network is asked to classify them and determine relationships.<br />
::'''PredCIs''': As above, except the classes are also provided. The only goal is to predict relationships.<br />
<br />
Further analysis into the accuracy, when looking at predicates individually, shows that the architecture is very sensitive to over-represented relationship predicates.<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Results - Part 2.PNG]]</div><br />
<br />
As shown in Figure 5, for many ground-truth predicates (those that do not appear often in the ground truth), the network does poorly. Even when allowed to propose 100 tuples, the network does not offer the predicate. Figure 4 simply observes the fact that certain sets of relationship predicates appear predominantly in a subset of slots. No general explanation has been offered for this behaviour.<br />
<br />
== Conclusion ==<br />
In conclusion, the paper offers a novel approach to that enables the extraction of image semantics while perpetually reasoning over the entire context of the image. Associative embeddings are used to connect object and predicate relationships, and parallel “slots” allow for multiple detections in one pixel. While this approach offers noticeable improvements in accuracy, it is clear that work needs to be done to account for the non-uniform distributions of relationships in the dataset.<br />
<br />
== Appendices ==<br />
<br />
'''Appendix 1: Sample Outputs'''<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Sample Pixel Graph Outputs.PNG]]</div><br />
<br />
'''Appendix 2: Stacked Hourglass Architecture'''<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Hourglass.PNG]]</div><br />
<br />
Although this goes beyond the focus of the paper, I would like to add a brief overview of the stacked hourglass architecture used to generate the heatmap. This architecture is unique in that it allows cyclical top-down, bottom-up inference and recombination of features. While most architectures focus on optimizing the bottom-up portion (reducing dimensionality), the stacked-hourglass gives the network more flexibility in how it generates a representation by allowing it to learn a series of down-sampling / up-sampling steps.<br />
<br />
== References ==<br />
1. Alejandro Newell and Jia Deng, “Pixels to Graphs by Associative Embedding,” in NIPS, 2017<br />
<br />
2. Alejandro Newell, Kaiyu Yang, and Jia Deng. Stacked Hourglass Networks for Human Pose Estimation. ECCV, 2016</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Pixels_to_Graphs_by_Associative_Embedding&diff=37172Pixels to Graphs by Associative Embedding2018-10-24T02:45:25Z<p>Gchalato: </p>
<hr />
<div>== Introduction == <br />
The paper presents a novel approach to proposing a scene graph. A scene graph, as it relates to an image, is a graph with a vertex that represents each object identified in the image and an edge that represents relationships between the objects. <br />
<br />
An example of a scene graph:<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Scene Graph.PNG]]</div><br />
<br />
Current state-of-the-art techniques break down the construction of scene graphs by first identifying objects, and then predicting the edges for any given pair of identified objects. This paper introduces an architecture that defines the entire graph directly from the image, enabling the network to reason across the entirety of the image to understand relationships, as opposed to only predicting relationships using object labels. <br />
<br />
A key concern, given that the new architecture produces both vertices (objects) and edges (relationships) , is connecting the two. Specifically, the output of the network is some set of relationships E, and some set of Vertices V. The network needs to also output the “source” and “destination” of each relationship, so that the final graph can be formed. In the image above, for example, the network would also need to tell us that “holding” comes from “person” and goes to “Frisbee”. To do this, the paper uses associative embeddings. Specifically, the network outputs a particular “embedding vector” for each vertex, as well as a “source embedding” and “destination embedding” for each relationship. A final post-processing step finds the vertex embedding closest to each of the source / destination embeddings of each relationship and in this way assigns the edges to pairs of vertices.<br />
<br />
== The Architecture: == <br />
: '''1. Detecting Graph Elements'''<br />
<br />
Given an image of dimensions h x w, a stacked hourglass (Appendix 2) is used to generate a h x w x f representation of the image. It should be noted that the dimension of the output (which is non-trainable), to consider. Specifically, we need to have a resolution large enough to minimize the number of pixels with multiple detections while also being small enough to ensure that each 1 x 1 x f vector still contains the information needed for subsequent inference.<br />
<br />
A 1x1 convolution and sigmoid activation is performed on this result to generate a heatmap (one for objects and one for relationships, using separately determined convolutions). The value at a given pixel can be interpreted as the likelihood of detection at that particular pixel in the original image. <br />
<br />
In order to claim that there is an element at some pixel, we need to have some likelihood threshold. Then, if a given pixel in the map has a value >= the threshold, we claim that there is an element at that pixel. This threshold is calculated by using binary cross-entropy loss on the final values in the heatmap. Values with likelihoods greater than p-hat will be considered element detections. <br />
<br />
Finally, for each element that we detected, we extract the 1 x 1 x f feature vector. This is then used as an input to a set of FFNN, where we have a separate network for each characteristic of interest. The following image summarizes the process.<br />
<br />
<br />
[[File:Extraction Process.PNG|center]]<br />
<br />
:'''2. Connecting Elements with Associative Embeddings'''<br />
As explained earlier, to construct the scene graph, we need to know the source and destination of each edge. This is done through associative embeddings. <br />
<br />
First, let us define an embedding hi ϵ Rd produced for some vector i, and let us assume that we have n object detections in a particular image. Now, define hik, for k = 1 to Ki (where Ki is the number of edges in the graph with a vertex at vertex i) as the embedding associated with an edge that touches vertex i. We define two loss functions on these sets.<br />
<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Loss 1.PNG]]</div><br />
<br />
The goal of Lpull is minimize the squared differences between the embedding of a given vertex and the embedding of an edge that references said vertex.<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Loss 2.PNG]]</div><br />
<br />
On the other hand, minimizing Lpush implies assigning embeddings to vertices that are as far apart as possible. The further apart they are, the lower the output of max becomes, until eventually it reaches 0. Here, m is just a constant. In the paper, the values used were m = 8 and d = 8 (that is, 8D embeddings). Combining these two loss functions (and weighing them equally), accomplishes the task of predicting embeddings such that vertices are differentiated, but the embedding of a vertex is most similar to the vertex it references.<br />
<br />
:'''3. Support for Overlapping Detections'''<br />
An obvious concern is how the network would operate if there was more than one detection (be it object or relationship), in a given pixel. For example, detection of “shirt” and “person” may be centered at the exact same pixel. To account for this, the architecture is modified to allow for “slots” at each pixel. Specifically, so detections of objects are allowed at a particular pixel, while sr relationship detections are allowed at a given pixel. <br />
<br />
In order to allow for this, some changes are required after the feature extraction step. Specifically, we now use the 1x1xf vector as the input for so (or sr) different sets of 4 FFNNs, where the output is as shown in figure 2, but with the final FFNN outputting the probability of a detection existing in that particular slot, at that particular pixel. This new network is trained exclusively on whether or not a detection has been made in that slot, and, in prediction, is used to determine the number of slots to output at a given pixel. It is critical to note that this each of these so (or sr) sets of FFNNs share absolutely no weights. And each is trained for detection in its assigned slot.<br />
<br />
It is important to note that that this implies a change in the training procedure. We now have so (or sr) different predictions (be it class, or class + bounding box), that we need to match with our set of ground truth detections at a given pixel. Without this step, we would not be able to assign a value to the error for that sample. To do this, we match a one-hot encoded vector of the ground-truth class and bounding box anchor (the reference vector), and then match them with the so (or sr) outputs provided at a given pixel. The Hungarian method is used to ensure maximum matching between the outputs and the reference method while ensuring we do not assign the same detection to multiple slots. <br />
<br />
<br />
==Results==<br />
A quick note on notation: R@50 indicates what percentage of ground-truth subject-predicate-object tuples appeared in a proposal of 50 such tuples. Since R@100 offers more possibilities, it will necessarily be higher. The 6.7, for example, indicates that 6.7% of the ground truth tuples appeared in the proposals of the network. <br />
<br />
The authors tested the network against two other architectures designed to develop semantic understanding of images. For this, they used the Visual Genome dataset, with so = 3 and sr = 6. Overall, the new architecture vastly outperformed past models. The results were as follows:<br />
<br />
The table can be interpreted as follows:<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Results Table.PNG]]</div><br />
<br />
::'''SGGen (no RPN)''': Given a particular image, without the use of Region Proposal networks, the accuracy of the proposed scene graph. No class predictions are provided.<br />
::'''SGGen (with RPN)''': Same as above, except the output of the Region Proposal Network is used to enhance the input of a given image. No class predictions are provided.<br />
::'''SGCIs''': Ground-truth object bounding boxes are provided. The network is asked to classify them and determine relationships.<br />
::'''PredCIs''': As above, except the classes are also provided. The only goal is to predict relationships.<br />
<br />
Further analysis into the accuracy, when looking at predicates individually, shows that the architecture is very sensitive to over-represented relationship predicates.<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Results - Part 2.PNG]]</div><br />
<br />
As shown in Figure 5, for many ground-truth predicates (those that do not appear often in the ground truth), the network does poorly. Even when allowed to propose 100 tuples, the network does not offer the predicate. Figure 4 simply observes the fact that certain sets of relationship predicates appear predominantly in a subset of slots. No general explanation has been offered for this behaviour.<br />
<br />
== Conclusion ==<br />
In conclusion, the paper offers a novel approach to that enables the extraction of image semantics while perpetually reasoning over the entire context of the image. Associative embeddings are used to connect object and predicate relationships, and parallel “slots” allow for multiple detections in one pixel. While this approach offers noticeable improvements in accuracy, it is clear that work needs to be done to account for the non-uniform distributions of relationships in the dataset.<br />
<br />
== Appendices ==<br />
<br />
'''Appendix 1: Sample Outputs'''<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Sample Pixel Graph Outputs.PNG]]</div><br />
<br />
'''Appendix 2: Stacked Hourglass Architecture'''<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Hourglass.PNG]]</div><br />
<br />
Although this goes beyond the focus of the paper, I would like to add a brief overview of the stacked hourglass architecture used to generate the heatmap. This architecture is unique in that it allows cyclical top-down, bottom-up inference and recombination of features. While most architectures focus on optimizing the bottom-up portion (reducing dimensionality), the stacked-hourglass gives the network more flexibility in how it generates a representation by allowing it to learn a series of down-sampling / up-sampling steps.<br />
<br />
== References ==<br />
1. Alejandro Newell and Jia Deng, “Pixels to Graphs by Associative Embedding,” in NIPS, 2017<br />
2. Alejandro Newell, Kaiyu Yang, and Jia Deng. Stacked Hourglass Networks for Human Pose Estimation. ECCV, 2016</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Pixels_to_Graphs_by_Associative_Embedding&diff=37171Pixels to Graphs by Associative Embedding2018-10-24T02:40:07Z<p>Gchalato: </p>
<hr />
<div>== Introduction == <br />
The paper presents a novel approach to proposing a scene graph. A scene graph, as it relates to an image, is a graph with a vertex that represents each object identified in the image and an edge that represents relationships between the objects. <br />
<br />
An example of a scene graph:<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Scene Graph.PNG]]</div><br />
<br />
Current state-of-the-art techniques break down the construction of scene graphs by first identifying objects, and then predicting the edges for any given pair of identified objects. This paper introduces an architecture that defines the entire graph directly from the image, enabling the network to reason across the entirety of the image to understand relationships, as opposed to only predicting relationships using object labels. <br />
A key concern, given that the new architecture produces both vertices (objects) and edges (relationships) , is connecting the two. Specifically, the output of the network is some set of relationships E, and some set of Vertices V. The network needs to also output the “source” and “destination” of each relationship, so that the final graph can be formed. In the image above, for example, the network would also need to tell us that “holding” comes from “person” and goes to “Frisbee”. To do this, the paper uses associative embeddings. Specifically, the network outputs a particular “embedding vector” for each vertex, as well as a “source embedding” and “destination embedding” for each relationship. A final post-processing step finds the vertex embedding closest to each of the source / destination embeddings of each relationship and in this way assigns the edges to pairs of vertices.<br />
<br />
== The Architecture: == <br />
: '''1. Detecting Graph Elements'''<br />
<br />
Given an image of dimensions h x w, a stacked hourglass (Appendix 2) is used to generate a h x w x f representation of the image. It should be noted that the dimension of the output (which is non-trainable), to consider. Specifically, we need to have a resolution large enough to minimize the number of pixels with multiple detections while also being small enough to ensure that each 1 x 1 x f vector still contains the information needed for subsequent inference.<br />
A 1x1 convolution and sigmoid activation is performed on this result to generate a heatmap (one for objects and one for relationships, using separately determined convolutions). The value at a given pixel can be interpreted as the likelihood of detection at that particular pixel in the original image. <br />
In order to claim that there is an element at some pixel, we need to have some likelihood threshold. Then, if a given pixel in the map has a value >= the threshold, we claim that there is an element at that pixel. This threshold is calculated by using binary cross-entropy loss on the final values in the heatmap. Values with likelihoods greater than p-hat will be considered element detections. <br />
Finally, for each element that we detected, we extract the 1 x 1 x f feature vector. This is then used as an input to a set of FFNN, where we have a separate network for each characteristic of interest. The following image summarizes the process.<br />
<br />
<br />
[[File:Extraction Process.PNG|center]]<br />
<br />
:'''2. Connecting Elements with Associative Embeddings'''<br />
As explained earlier, to construct the scene graph, we need to know the source and destination of each edge. This is done through associative embeddings. <br />
First, let us define an embedding hi ϵ Rd produced for some vector i, and let us assume that we have n object detections in a particular image. Now, define hik, for k = 1 to Ki (where Ki is the number of edges in the graph with a vertex at vertex i) as the embedding associated with an edge that touches vertex i. We define two loss functions on these sets.<br />
<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Loss 1.PNG]]</div><br />
<br />
The goal of Lpull is minimize the squared differences between the embedding of a given vertex and the embedding of an edge that references said vertex.<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Loss 2.PNG]]</div><br />
<br />
On the other hand, minimizing Lpush implies assigning embeddings to vertices that are as far apart as possible. The further apart they are, the lower the output of max becomes, until eventually it reaches 0. Here, m is just a constant. In the paper, the values used were m = 8 and d = 8 (that is, 8D embeddings). Combining these two loss functions (and weighing them equally), accomplishes the task of predicting embeddings such that vertices are differentiated, but the embedding of a vertex is most similar to the vertex it references.<br />
<br />
:'''3. Support for Overlapping Detections'''<br />
An obvious concern is how the network would operate if there was more than one detection (be it object or relationship), in a given pixel. For example, detection of “shirt” and “person” may be centered at the exact same pixel. To account for this, the architecture is modified to allow for “slots” at each pixel. Specifically, so detections of objects are allowed at a particular pixel, while sr relationship detections are allowed at a given pixel. <br />
In order to allow for this, some changes are required after the feature extraction step. Specifically, we now use the 1x1xf vector as the input for so (or sr) different sets of 4 FFNNs, where the output is as shown in figure 2, but with the final FFNN outputting the probability of a detection existing in that particular slot, at that particular pixel. This new network is trained exclusively on whether or not a detection has been made in that slot, and, in prediction, is used to determine the number of slots to output at a given pixel. It is critical to note that this each of these so (or sr) sets of FFNNs share absolutely no weights. And each is trained for detection in its assigned slot.<br />
It is important to note that that this implies a change in the training procedure. We now have so (or sr) different predictions (be it class, or class + bounding box), that we need to match with our set of ground truth detections at a given pixel. Without this step, we would not be able to assign a value to the error for that sample. To do this, we match a one-hot encoded vector of the ground-truth class and bounding box anchor (the reference vector), and then match them with the so (or sr) outputs provided at a given pixel. The Hungarian method is used to ensure maximum matching between the outputs and the reference method while ensuring we do not assign the same detection to multiple slots. <br />
<br />
<br />
==Results==<br />
A quick note on notation: R@50 indicates what percentage of ground-truth subject-predicate-object tuples appeared in a proposal of 50 such tuples. Since R@100 offers more possibilities, it will necessarily be higher. The 6.7, for example, indicates that 6.7% of the ground truth tuples appeared in the proposals of the network. <br />
<br />
The authors tested the network against two other architectures designed to develop semantic understanding of images. For this, they used the Visual Genome dataset, with so = 3 and sr = 6. Overall, the new architecture vastly outperformed past models. The results were as follows:<br />
<br />
The table can be interpreted as follows:<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Results Table.PNG]]</div><br />
<br />
::'''SGGen (no RPN)''': Given a particular image, without the use of Region Proposal networks, the accuracy of the proposed scene graph. No class predictions are provided.<br />
::'''SGGen (with RPN)''': Same as above, except the output of the Region Proposal Network is used to enhance the input of a given image. No class predictions are provided.<br />
::'''SGCIs''': Ground-truth object bounding boxes are provided. The network is asked to classify them and determine relationships.<br />
::'''PredCIs''': As above, except the classes are also provided. The only goal is to predict relationships.<br />
<br />
Further analysis into the accuracy, when looking at predicates individually, shows that the architecture is very sensitive to over-represented relationship predicates.<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Results - Part 2.PNG]]</div><br />
<br />
As shown in Figure 5, for many ground-truth predicates (those that do not appear often in the ground truth), the network does poorly. Even when allowed to propose 100 tuples, the network does not offer the predicate. Figure 4 simply observes the fact that certain sets of relationship predicates appear predominantly in a subset of slots. No general explanation has been offered for this behaviour.<br />
<br />
== Conclusion ==<br />
In conclusion, the paper offers a novel approach to that enables the extraction of image semantics while perpetually reasoning over the entire context of the image. Associative embeddings are used to connect object and predicate relationships, and parallel “slots” allow for multiple detections in one pixel. While this approach offers noticeable improvements in accuracy, it is clear that work needs to be done to account for the non-uniform distributions of relationships in the dataset.<br />
<br />
== Appendices ==<br />
<br />
'''Appendix 1: Sample Outputs'''<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Sample Pixel Graph Outputs.PNG]]</div><br />
<br />
'''Appendix 2: Stacked Hourglass Architecture'''<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Hourglass.PNG]]</div><br />
<br />
Although this goes beyond the focus of the paper, I would like to add a brief overview of the stacked hourglass architecture used to generate the heatmap. This architecture is unique in that it allows cyclical top-down, bottom-up inference and recombination of features. While most architectures focus on optimizing the bottom-up portion (reducing dimensionality), the stacked-hourglass gives the network more flexibility in how it generates a representation by allowing it to learn a series of down-sampling / up-sampling steps.<br />
<br />
== References ==<br />
1. Alejandro Newell and Jia Deng, “Pixels to Graphs by Associative Embedding,” in NIPS, 2017<br />
2. Alejandro Newell, Kaiyu Yang, and Jia Deng. Stacked Hourglass Networks for Human Pose Estimation. ECCV, 2016</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Pixels_to_Graphs_by_Associative_Embedding&diff=37170Pixels to Graphs by Associative Embedding2018-10-24T02:36:10Z<p>Gchalato: /* Results */</p>
<hr />
<div>== Introduction == <br />
The paper presents a novel approach to proposing a scene graph. A scene graph, as it relates to an image, is a graph with a vertex that represents each object identified in the image and an edge that represents relationships between the objects. <br />
<br />
An example of a scene graph:<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Scene Graph.PNG]]</div><br />
<br />
Current state-of-the-art techniques break down the construction of scene graphs by first identifying objects, and then predicting the edges for any given pair of identified objects. This paper introduces an architecture that defines the entire graph directly from the image, enabling the network to reason across the entirety of the image to understand relationships, as opposed to only predicting relationships using object labels. <br />
A key concern, given that the new architecture produces both vertices (objects) and edges (relationships) , is connecting the two. Specifically, the output of the network is some set of relationships E, and some set of Vertices V. The network needs to also output the “source” and “destination” of each relationship, so that the final graph can be formed. In the image above, for example, the network would also need to tell us that “holding” comes from “person” and goes to “Frisbee”. To do this, the paper uses associative embeddings. Specifically, the network outputs a particular “embedding vector” for each vertex, as well as a “source embedding” and “destination embedding” for each relationship. A final post-processing step finds the vertex embedding closest to each of the source / destination embeddings of each relationship and in this way assigns the edges to pairs of vertices.<br />
<br />
== The Architecture: == <br />
: '''1. Detecting Graph Elements'''<br />
<br />
Given an image of dimensions h x w, a stacked hourglass (Appendix 2) is used to generate a h x w x f representation of the image. It should be noted that the dimension of the output (which is non-trainable), to consider. Specifically, we need to have a resolution large enough to minimize the number of pixels with multiple detections while also being small enough to ensure that each 1 x 1 x f vector still contains the information needed for subsequent inference.<br />
A 1x1 convolution and sigmoid activation is performed on this result to generate a heatmap (one for objects and one for relationships, using separately determined convolutions). The value at a given pixel can be interpreted as the likelihood of detection at that particular pixel in the original image. <br />
In order to claim that there is an element at some pixel, we need to have some likelihood threshold. Then, if a given pixel in the map has a value >= the threshold, we claim that there is an element at that pixel. This threshold is calculated by using binary cross-entropy loss on the final values in the heatmap. Values with likelihoods greater than p-hat will be considered element detections. <br />
Finally, for each element that we detected, we extract the 1 x 1 x f feature vector. This is then used as an input to a set of FFNN, where we have a separate network for each characteristic of interest. The following image summarizes the process.<br />
<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Extraction Process.PNG]]</div><br />
<br />
:'''2. Connecting Elements with Associative Embeddings'''<br />
As explained earlier, to construct the scene graph, we need to know the source and destination of each edge. This is done through associative embeddings. <br />
First, let us define an embedding hi ϵ Rd produced for some vector i, and let us assume that we have n object detections in a particular image. Now, define hik, for k = 1 to Ki (where Ki is the number of edges in the graph with a vertex at vertex i) as the embedding associated with an edge that touches vertex i. We define two loss functions on these sets.<br />
<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Loss 1.PNG]]</div><br />
<br />
The goal of Lpull is minimize the squared differences between the embedding of a given vertex and the embedding of an edge that references said vertex.<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Loss 2.PNG]]</div><br />
<br />
On the other hand, minimizing Lpush implies assigning embeddings to vertices that are as far apart as possible. The further apart they are, the lower the output of max becomes, until eventually it reaches 0. Here, m is just a constant. In the paper, the values used were m = 8 and d = 8 (that is, 8D embeddings). Combining these two loss functions (and weighing them equally), accomplishes the task of predicting embeddings such that vertices are differentiated, but the embedding of a vertex is most similar to the vertex it references.<br />
<br />
:'''3. Support for Overlapping Detections'''<br />
An obvious concern is how the network would operate if there was more than one detection (be it object or relationship), in a given pixel. For example, detection of “shirt” and “person” may be centered at the exact same pixel. To account for this, the architecture is modified to allow for “slots” at each pixel. Specifically, so detections of objects are allowed at a particular pixel, while sr relationship detections are allowed at a given pixel. <br />
In order to allow for this, some changes are required after the feature extraction step. Specifically, we now use the 1x1xf vector as the input for so (or sr) different sets of 4 FFNNs, where the output is as shown in figure 2, but with the final FFNN outputting the probability of a detection existing in that particular slot, at that particular pixel. This new network is trained exclusively on whether or not a detection has been made in that slot, and, in prediction, is used to determine the number of slots to output at a given pixel. It is critical to note that this each of these so (or sr) sets of FFNNs share absolutely no weights. And each is trained for detection in its assigned slot.<br />
It is important to note that that this implies a change in the training procedure. We now have so (or sr) different predictions (be it class, or class + bounding box), that we need to match with our set of ground truth detections at a given pixel. Without this step, we would not be able to assign a value to the error for that sample. To do this, we match a one-hot encoded vector of the ground-truth class and bounding box anchor (the reference vector), and then match them with the so (or sr) outputs provided at a given pixel. The Hungarian method is used to ensure maximum matching between the outputs and the reference method while ensuring we do not assign the same detection to multiple slots. <br />
<br />
<br />
==Results==<br />
A quick note on notation: R@50 indicates what percentage of ground-truth subject-predicate-object tuples appeared in a proposal of 50 such tuples. Since R@100 offers more possibilities, it will necessarily be higher. The 6.7, for example, indicates that 6.7% of the ground truth tuples appeared in the proposals of the network. <br />
<br />
The authors tested the network against two other architectures designed to develop semantic understanding of images. For this, they used the Visual Genome dataset, with so = 3 and sr = 6. Overall, the new architecture vastly outperformed past models. The results were as follows:<br />
<br />
The table can be interpreted as follows:<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Results Table.PNG]]</div><br />
<br />
::'''SGGen (no RPN)''': Given a particular image, without the use of Region Proposal networks, the accuracy of the proposed scene graph. No class predictions are provided.<br />
::'''SGGen (with RPN)''': Same as above, except the output of the Region Proposal Network is used to enhance the input of a given image. No class predictions are provided.<br />
::'''SGCIs''': Ground-truth object bounding boxes are provided. The network is asked to classify them and determine relationships.<br />
::'''PredCIs''': As above, except the classes are also provided. The only goal is to predict relationships.<br />
<br />
Further analysis into the accuracy, when looking at predicates individually, shows that the architecture is very sensitive to over-represented relationship predicates.<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Results - Part 2.PNG]]</div><br />
<br />
As shown in Figure 5, for many ground-truth predicates (those that do not appear often in the ground truth), the network does poorly. Even when allowed to propose 100 tuples, the network does not offer the predicate. Figure 4 simply observes the fact that certain sets of relationship predicates appear predominantly in a subset of slots. No general explanation has been offered for this behaviour.<br />
<br />
== Conclusion ==<br />
In conclusion, the paper offers a novel approach to that enables the extraction of image semantics while perpetually reasoning over the entire context of the image. Associative embeddings are used to connect object and predicate relationships, and parallel “slots” allow for multiple detections in one pixel. While this approach offers noticeable improvements in accuracy, it is clear that work needs to be done to account for the non-uniform distributions of relationships in the dataset.<br />
<br />
== Appendices ==<br />
<br />
'''Appendix 1: Sample Outputs'''<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Sample Pixel Graph Outputs.PNG]]</div><br />
<br />
'''Appendix 2: Stacked Hourglass Architecture'''<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Hourglass.PNG]]</div><br />
<br />
Although this goes beyond the focus of the paper, I would like to add a brief overview of the stacked hourglass architecture used to generate the heatmap. This architecture is unique in that it allows cyclical top-down, bottom-up inference and recombination of features. While most architectures focus on optimizing the bottom-up portion (reducing dimensionality), the stacked-hourglass gives the network more flexibility in how it generates a representation by allowing it to learn a series of down-sampling / up-sampling steps.<br />
<br />
== References ==<br />
1. Alejandro Newell and Jia Deng, “Pixels to Graphs by Associative Embedding,” in NIPS, 2017<br />
2. Alejandro Newell, Kaiyu Yang, and Jia Deng. Stacked Hourglass Networks for Human Pose Estimation. ECCV, 2016</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Pixels_to_Graphs_by_Associative_Embedding&diff=37169Pixels to Graphs by Associative Embedding2018-10-24T02:35:24Z<p>Gchalato: </p>
<hr />
<div>== Introduction == <br />
The paper presents a novel approach to proposing a scene graph. A scene graph, as it relates to an image, is a graph with a vertex that represents each object identified in the image and an edge that represents relationships between the objects. <br />
<br />
An example of a scene graph:<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Scene Graph.PNG]]</div><br />
<br />
Current state-of-the-art techniques break down the construction of scene graphs by first identifying objects, and then predicting the edges for any given pair of identified objects. This paper introduces an architecture that defines the entire graph directly from the image, enabling the network to reason across the entirety of the image to understand relationships, as opposed to only predicting relationships using object labels. <br />
A key concern, given that the new architecture produces both vertices (objects) and edges (relationships) , is connecting the two. Specifically, the output of the network is some set of relationships E, and some set of Vertices V. The network needs to also output the “source” and “destination” of each relationship, so that the final graph can be formed. In the image above, for example, the network would also need to tell us that “holding” comes from “person” and goes to “Frisbee”. To do this, the paper uses associative embeddings. Specifically, the network outputs a particular “embedding vector” for each vertex, as well as a “source embedding” and “destination embedding” for each relationship. A final post-processing step finds the vertex embedding closest to each of the source / destination embeddings of each relationship and in this way assigns the edges to pairs of vertices.<br />
<br />
== The Architecture: == <br />
: '''1. Detecting Graph Elements'''<br />
<br />
Given an image of dimensions h x w, a stacked hourglass (Appendix 2) is used to generate a h x w x f representation of the image. It should be noted that the dimension of the output (which is non-trainable), to consider. Specifically, we need to have a resolution large enough to minimize the number of pixels with multiple detections while also being small enough to ensure that each 1 x 1 x f vector still contains the information needed for subsequent inference.<br />
A 1x1 convolution and sigmoid activation is performed on this result to generate a heatmap (one for objects and one for relationships, using separately determined convolutions). The value at a given pixel can be interpreted as the likelihood of detection at that particular pixel in the original image. <br />
In order to claim that there is an element at some pixel, we need to have some likelihood threshold. Then, if a given pixel in the map has a value >= the threshold, we claim that there is an element at that pixel. This threshold is calculated by using binary cross-entropy loss on the final values in the heatmap. Values with likelihoods greater than p-hat will be considered element detections. <br />
Finally, for each element that we detected, we extract the 1 x 1 x f feature vector. This is then used as an input to a set of FFNN, where we have a separate network for each characteristic of interest. The following image summarizes the process.<br />
<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Extraction Process.PNG]]</div><br />
<br />
:'''2. Connecting Elements with Associative Embeddings'''<br />
As explained earlier, to construct the scene graph, we need to know the source and destination of each edge. This is done through associative embeddings. <br />
First, let us define an embedding hi ϵ Rd produced for some vector i, and let us assume that we have n object detections in a particular image. Now, define hik, for k = 1 to Ki (where Ki is the number of edges in the graph with a vertex at vertex i) as the embedding associated with an edge that touches vertex i. We define two loss functions on these sets.<br />
<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Loss 1.PNG]]</div><br />
<br />
The goal of Lpull is minimize the squared differences between the embedding of a given vertex and the embedding of an edge that references said vertex.<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Loss 2.PNG]]</div><br />
<br />
On the other hand, minimizing Lpush implies assigning embeddings to vertices that are as far apart as possible. The further apart they are, the lower the output of max becomes, until eventually it reaches 0. Here, m is just a constant. In the paper, the values used were m = 8 and d = 8 (that is, 8D embeddings). Combining these two loss functions (and weighing them equally), accomplishes the task of predicting embeddings such that vertices are differentiated, but the embedding of a vertex is most similar to the vertex it references.<br />
<br />
:'''3. Support for Overlapping Detections'''<br />
An obvious concern is how the network would operate if there was more than one detection (be it object or relationship), in a given pixel. For example, detection of “shirt” and “person” may be centered at the exact same pixel. To account for this, the architecture is modified to allow for “slots” at each pixel. Specifically, so detections of objects are allowed at a particular pixel, while sr relationship detections are allowed at a given pixel. <br />
In order to allow for this, some changes are required after the feature extraction step. Specifically, we now use the 1x1xf vector as the input for so (or sr) different sets of 4 FFNNs, where the output is as shown in figure 2, but with the final FFNN outputting the probability of a detection existing in that particular slot, at that particular pixel. This new network is trained exclusively on whether or not a detection has been made in that slot, and, in prediction, is used to determine the number of slots to output at a given pixel. It is critical to note that this each of these so (or sr) sets of FFNNs share absolutely no weights. And each is trained for detection in its assigned slot.<br />
It is important to note that that this implies a change in the training procedure. We now have so (or sr) different predictions (be it class, or class + bounding box), that we need to match with our set of ground truth detections at a given pixel. Without this step, we would not be able to assign a value to the error for that sample. To do this, we match a one-hot encoded vector of the ground-truth class and bounding box anchor (the reference vector), and then match them with the so (or sr) outputs provided at a given pixel. The Hungarian method is used to ensure maximum matching between the outputs and the reference method while ensuring we do not assign the same detection to multiple slots. <br />
<br />
<br />
==Results==<br />
A quick note on notation: R@50 indicates what percentage of ground-truth subject-predicate-object tuples appeared in a proposal of 50 such tuples. Since R@100 offers more possibilities, it will necessarily be higher. The 6.7, for example, indicates that 6.7% of the ground truth pairs appeared in the proposals of the network. <br />
<br />
The authors tested the network against two other architectures designed to develop semantic understanding of images. For this, they used the Visual Genome dataset, with so = 3 and sr = 6. Overall, the new architecture vastly outperformed past models. The results were as follows:<br />
<br />
The table can be interpreted as follows:<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Results Table.PNG]]</div><br />
<br />
::'''SGGen (no RPN)''': Given a particular image, without the use of Region Proposal networks, the accuracy of the proposed scene graph. No class predictions are provided.<br />
::'''SGGen (with RPN)''': Same as above, except the output of the Region Proposal Network is used to enhance the input of a given image. No class predictions are provided.<br />
::'''SGCIs''': Ground-truth object bounding boxes are provided. The network is asked to classify them and determine relationships.<br />
::'''PredCIs''': As above, except the classes are also provided. The only goal is to predict relationships.<br />
<br />
Further analysis into the accuracy, when looking at predicates individually, shows that the architecture is very sensitive to over-represented relationship predicates.<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Results - Part 2.PNG]]</div><br />
<br />
As shown in Figure 5, for many ground-truth predicates (those that do not appear often in the ground truth), the network does poorly. Even when allowed to propose 100 tuples, the network does not offer the predicate. Figure 4 simply observes the fact that certain sets of relationship predicates appear predominantly in a subset of slots. No general explanation has been offered for this behaviour.<br />
<br />
== Conclusion ==<br />
In conclusion, the paper offers a novel approach to that enables the extraction of image semantics while perpetually reasoning over the entire context of the image. Associative embeddings are used to connect object and predicate relationships, and parallel “slots” allow for multiple detections in one pixel. While this approach offers noticeable improvements in accuracy, it is clear that work needs to be done to account for the non-uniform distributions of relationships in the dataset.<br />
<br />
== Appendices ==<br />
<br />
'''Appendix 1: Sample Outputs'''<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Sample Pixel Graph Outputs.PNG]]</div><br />
<br />
'''Appendix 2: Stacked Hourglass Architecture'''<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Hourglass.PNG]]</div><br />
<br />
Although this goes beyond the focus of the paper, I would like to add a brief overview of the stacked hourglass architecture used to generate the heatmap. This architecture is unique in that it allows cyclical top-down, bottom-up inference and recombination of features. While most architectures focus on optimizing the bottom-up portion (reducing dimensionality), the stacked-hourglass gives the network more flexibility in how it generates a representation by allowing it to learn a series of down-sampling / up-sampling steps.<br />
<br />
== References ==<br />
1. Alejandro Newell and Jia Deng, “Pixels to Graphs by Associative Embedding,” in NIPS, 2017<br />
2. Alejandro Newell, Kaiyu Yang, and Jia Deng. Stacked Hourglass Networks for Human Pose Estimation. ECCV, 2016</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Pixels_to_Graphs_by_Associative_Embedding&diff=37168Pixels to Graphs by Associative Embedding2018-10-24T02:33:43Z<p>Gchalato: </p>
<hr />
<div>== Introduction == <br />
The paper presents a novel approach to proposing a scene graph. A scene graph, as it relates to an image, is a graph with a vertex that represents each object identified in the image and an edge that represents relationships between the objects. <br />
<br />
An example of a scene graph:<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Scene Graph.PNG]]</div><br />
<br />
Current state-of-the-art techniques break down the construction of scene graphs by first identifying objects, and then predicting the edges for any given pair of identified objects. This paper introduces an architecture that defines the entire graph directly from the image, enabling the network to reason across the entirety of the image to understand relationships, as opposed to only predicting relationships using object labels. <br />
A key concern, given that the new architecture produces both vertices (objects) and edges (relationships) , is connecting the two. Specifically, the output of the network is some set of relationships E, and some set of Vertices V. The network needs to also output the “source” and “destination” of each relationship, so that the final graph can be formed. In the image above, for example, the network would also need to tell us that “holding” comes from “person” and goes to “Frisbee”. To do this, the paper uses associative embeddings. Specifically, the network outputs a particular “embedding vector” for each vertex, as well as a “source embedding” and “destination embedding” for each relationship. A final post-processing step finds the vertex embedding closest to each of the source / destination embeddings of each relationship and in this way assigns the edges to pairs of vertices.<br />
<br />
== The Architecture: == <br />
: '''1. Detecting Graph Elements'''<br />
<br />
Given an image of dimensions h x w, a stacked hourglass (Appendix 2) is used to generate a h x w x f representation of the image. It should be noted that the dimension of the output (which is non-trainable), to consider. Specifically, we need to have a resolution large enough to minimize the number of pixels with multiple detections while also being small enough to ensure that each 1 x 1 x f vector still contains the information needed for subsequent inference.<br />
A 1x1 convolution and sigmoid activation is performed on this result to generate a heatmap (one for objects and one for relationships, using separately determined convolutions). The value at a given pixel can be interpreted as the likelihood of detection at that particular pixel in the original image. <br />
In order to claim that there is an element at some pixel, we need to have some likelihood threshold. Then, if a given pixel in the map has a value >= the threshold, we claim that there is an element at that pixel. This threshold is calculated by using binary cross-entropy loss on the final values in the heatmap. Values with likelihoods greater than p-hat will be considered element detections. <br />
Finally, for each element that we detected, we extract the 1 x 1 x f feature vector. This is then used as an input to a set of FFNN, where we have a separate network for each characteristic of interest. The following image summarizes the process.<br />
<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Extraction Process.PNG]]</div><br />
<br />
:'''2. Connecting Elements with Associative Embeddings'''<br />
As explained earlier, to construct the scene graph, we need to know the source and destination of each edge. This is done through associative embeddings. <br />
First, let us define an embedding hi ϵ Rd produced for some vector i, and let us assume that we have n object detections in a particular image. Now, define hik, for k = 1 to Ki (where Ki is the number of edges in the graph with a vertex at vertex i) as the embedding associated with an edge that touches vertex i. We define two loss functions on these sets.<br />
<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Loss 1.PNG]]</div><br />
<br />
The goal of Lpull is minimize the squared differences between the embedding of a given vertex and the embedding of an edge that references said vertex.<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Loss 2.PNG]]</div><br />
<br />
On the other hand, minimizing Lpush implies assigning embeddings to vertices that are as far apart as possible. The further apart they are, the lower the output of max becomes, until eventually it reaches 0. Here, m is just a constant. In the paper, the values used were m = 8 and d = 8 (that is, 8D embeddings). Combining these two loss functions (and weighing them equally), accomplishes the task of predicting embeddings such that vertices are differentiated, but the embedding of a vertex is most similar to the vertex it references.<br />
<br />
:'''3. Support for Overlapping Detections'''<br />
An obvious concern is how the network would operate if there was more than one detection (be it object or relationship), in a given pixel. For example, detection of “shirt” and “person” may be centered at the exact same pixel. To account for this, the architecture is modified to allow for “slots” at each pixel. Specifically, so detections of objects are allowed at a particular pixel, while sr relationship detections are allowed at a given pixel. <br />
In order to allow for this, some changes are required after the feature extraction step. Specifically, we now use the 1x1xf vector as the input for so (or sr) different sets of 4 FFNNs, where the output is as shown in figure 2, but with the final FFNN outputting the probability of a detection existing in that particular slot, at that particular pixel. This new network is trained exclusively on whether or not a detection has been made in that slot, and, in prediction, is used to determine the number of slots to output at a given pixel. It is critical to note that this each of these so (or sr) sets of FFNNs share absolutely no weights. And each is trained for detection in its assigned slot.<br />
It is important to note that that this implies a change in the training procedure. We now have so (or sr) different predictions (be it class, or class + bounding box), that we need to match with our set of ground truth detections at a given pixel. Without this step, we would not be able to assign a value to the error for that sample. To do this, we match a one-hot encoded vector of the ground-truth class and bounding box anchor (the reference vector), and then match them with the so (or sr) outputs provided at a given pixel. The Hungarian method is used to ensure maximum matching between the outputs and the reference method while ensuring we do not assign the same detection to multiple slots. <br />
<br />
<br />
==Results==<br />
A quick note on notation: R@50 indicates what percentage of ground-truth subject-predicate-object tuples appeared in a proposal of 50 such tuples. Since R@100 offers more possibilities, it will necessarily be higher. The 6.7, for example, indicates that 6.7% of the ground truth pairs appeared in the proposals of the network. <br />
<br />
The authors tested the network against two other architectures designed to develop semantic understanding of images. For this, they used the Visual Genome dataset, with so = 3 and sr = 6. Overall, the new architecture vastly outperformed past models. The results were as follows:<br />
<br />
The table can be interpreted as follows:<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Results Table.PNG]]</div><br />
<br />
::'''SGGen (no RPN)''': Given a particular image, without the use of Region Proposal networks, the accuracy of the proposed scene graph. No class predictions are provided.<br />
::'''SGGen (with RPN)''': Same as above, except the output of the Region Proposal Network is used to enhance the input of a given image. No class predictions are provided.<br />
::'''SGCIs''': Ground-truth object bounding boxes are provided. The network is asked to classify them and determine relationships.<br />
::'''PredCIs''': As above, except the classes are also provided. The only goal is to predict relationships.<br />
<br />
Further analysis into the accuracy, when looking at predicates individually, shows that the architecture is very sensitive to over-represented relationship predicates.<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Results - Part 2.PNG]]</div><br />
<br />
As shown in Figure 5, for many ground-truth predicates (those that do not appear often in the ground truth), the network does poorly. Even when allowed to propose 100 tuples, the network does not offer the predicate. Figure 4 simply observes the fact that certain sets of relationship predicates appear predominantly in a subset of slots. No general explanation has been offered for this behaviour.<br />
<br />
== Conclusion ==<br />
In conclusion, the paper offers a novel approach to that enables the extraction of image semantics while perpetually reasoning over the entire context of the image. Associative embeddings are used to connect object and predicate relationships, and parallel “slots” allow for multiple detections in one pixel. While this approach offers noticeable improvements in accuracy, it is clear that work needs to be done to account for the non-uniform distributions of relationships in the dataset.<br />
<br />
== Appendices ==<br />
<br />
Appendix 1: Sample Outputs<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Sample Pixel Graph Outputs.PNG]]</div><br />
<br />
Appendix 2: Stacked Hourglass Architecture<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Hourglass.PNG]]</div><br />
<br />
Although this goes beyond the focus of the paper, I would like to add a brief overview of the stacked hourglass architecture used to generate the heatmap. This architecture is unique in that it allows cyclical top-down, bottom-up inference and recombination of features. While most architectures focus on optimizing the bottom-up portion (reducing dimensionality), the stacked-hourglass gives the network more flexibility in how it generates a representation by allowing it to learn a series of down-sampling / up-sampling steps.<br />
<br />
== References ==<br />
1. Alejandro Newell and Jia Deng, “Pixels to Graphs by Associative Embedding,” in NIPS, 2017<br />
2. Alejandro Newell, Kaiyu Yang, and Jia Deng. Stacked Hourglass Networks for Human Pose Estimation. ECCV, 2016</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Pixels_to_Graphs_by_Associative_Embedding&diff=37167Pixels to Graphs by Associative Embedding2018-10-24T02:30:28Z<p>Gchalato: /* Introduction */</p>
<hr />
<div>== Introduction == <br />
The paper presents a novel approach to proposing a scene graph. A scene graph, as it relates to an image, is a graph with a vertex that represents each object identified in the image and an edge that represents relationships between the objects. <br />
<br />
An example of a scene graph:<br />
<br />
<div class="center" style="width: auto; margin-left: auto; margin-right: auto;">[[File:Scene Graph.PNG]]</div><br />
<br />
Current state-of-the-art techniques break down the construction of scene graphs by first identifying objects, and then predicting the edges for any given pair of identified objects. This paper introduces an architecture that defines the entire graph directly from the image, enabling the network to reason across the entirety of the image to understand relationships, as opposed to only predicting relationships using object labels. <br />
A key concern, given that the new architecture produces both vertices (objects) and edges (relationships) , is connecting the two. Specifically, the output of the network is some set of relationships E, and some set of Vertices V. The network needs to also output the “source” and “destination” of each relationship, so that the final graph can be formed. In the image above, for example, the network would also need to tell us that “holding” comes from “person” and goes to “Frisbee”. To do this, the paper uses associative embeddings. Specifically, the network outputs a particular “embedding vector” for each vertex, as well as a “source embedding” and “destination embedding” for each relationship. A final post-processing step finds the vertex embedding closest to each of the source / destination embeddings of each relationship and in this way assigns the edges to pairs of vertices.<br />
<br />
== The Architecture: == <br />
: '''1. Detecting Graph Elements'''<br />
<br />
Given an image of dimensions h x w, a stacked hourglass (Appendix 2) is used to generate a h x w x f representation of the image. It should be noted that the dimension of the output (which is non-trainable), to consider. Specifically, we need to have a resolution large enough to minimize the number of pixels with multiple detections while also being small enough to ensure that each 1 x 1 x f vector still contains the information needed for subsequent inference.<br />
A 1x1 convolution and sigmoid activation is performed on this result to generate a heatmap (one for objects and one for relationships, using separately determined convolutions). The value at a given pixel can be interpreted as the likelihood of detection at that particular pixel in the original image. <br />
In order to claim that there is an element at some pixel, we need to have some likelihood threshold. Then, if a given pixel in the map has a value >= the threshold, we claim that there is an element at that pixel. This threshold is calculated by using binary cross-entropy loss on the final values in the heatmap. Values with likelihoods greater than p-hat will be considered element detections. <br />
Finally, for each element that we detected, we extract the 1 x 1 x f feature vector. This is then used as an input to a set of FFNN, where we have a separate network for each characteristic of interest. The following image summarizes the process.<br />
<br />
:'''2. Connecting Elements with Associative Embeddings'''<br />
As explained earlier, to construct the scene graph, we need to know the source and destination of each edge. This is done through associative embeddings. <br />
First, let us define an embedding hi ϵ Rd produced for some vector i, and let us assume that we have n object detections in a particular image. Now, define hik, for k = 1 to Ki (where Ki is the number of edges in the graph with a vertex at vertex i) as the embedding associated with an edge that touches vertex i. We define two loss functions on these sets.<br />
<br />
The goal of Lpull is minimize the squared differences between the embedding of a given vertex and the embedding of an edge that references said vertex.<br />
<br />
On the other hand, minimizing Lpush implies assigning embeddings to vertices that are as far apart as possible. The further apart they are, the lower the output of max becomes, until eventually it reaches 0. Here, m is just a constant. In the paper, the values used were m = 8 and d = 8 (that is, 8D embeddings). Combining these two loss functions (and weighing them equally), accomplishes the task of predicting embeddings such that vertices are differentiated, but the embedding of a vertex is most similar to the vertex it references.<br />
<br />
:'''3. Support for Overlapping Detections'''<br />
An obvious concern is how the network would operate if there was more than one detection (be it object or relationship), in a given pixel. For example, detection of “shirt” and “person” may be centered at the exact same pixel. To account for this, the architecture is modified to allow for “slots” at each pixel. Specifically, so detections of objects are allowed at a particular pixel, while sr relationship detections are allowed at a given pixel. <br />
In order to allow for this, some changes are required after the feature extraction step. Specifically, we now use the 1x1xf vector as the input for so (or sr) different sets of 4 FFNNs, where the output is as shown in figure 2, but with the final FFNN outputting the probability of a detection existing in that particular slot, at that particular pixel. This new network is trained exclusively on whether or not a detection has been made in that slot, and, in prediction, is used to determine the number of slots to output at a given pixel. It is critical to note that this each of these so (or sr) sets of FFNNs share absolutely no weights. And each is trained for detection in its assigned slot.<br />
It is important to note that that this implies a change in the training procedure. We now have so (or sr) different predictions (be it class, or class + bounding box), that we need to match with our set of ground truth detections at a given pixel. Without this step, we would not be able to assign a value to the error for that sample. To do this, we match a one-hot encoded vector of the ground-truth class and bounding box anchor (the reference vector), and then match them with the so (or sr) outputs provided at a given pixel. The Hungarian method is used to ensure maximum matching between the outputs and the reference method while ensuring we do not assign the same detection to multiple slots. <br />
<br />
<br />
==Results==<br />
A quick note on notation: R@50 indicates what percentage of ground-truth subject-predicate-object tuples appeared in a proposal of 50 such tuples. Since R@100 offers more possibilities, it will necessarily be higher. The 6.7, for example, indicates that 6.7% of the ground truth pairs appeared in the proposals of the network. <br />
<br />
The authors tested the network against two other architectures designed to develop semantic understanding of images. For this, they used the Visual Genome dataset, with so = 3 and sr = 6. Overall, the new architecture vastly outperformed past models. The results were as follows:<br />
<br />
The table can be interpreted as follows:<br />
<br />
::'''SGGen (no RPN)''': Given a particular image, without the use of Region Proposal networks, the accuracy of the proposed scene graph. No class predictions are provided.<br />
::'''SGGen (with RPN)''': Same as above, except the output of the Region Proposal Network is used to enhance the input of a given image. No class predictions are provided.<br />
::'''SGCIs''': Ground-truth object bounding boxes are provided. The network is asked to classify them and determine relationships.<br />
::'''PredCIs''': As above, except the classes are also provided. The only goal is to predict relationships.<br />
<br />
Further analysis into the accuracy, when looking at predicates individually, shows that the architecture is very sensitive to over-represented relationship predicates.<br />
<br />
As shown in Figure 5, for many ground-truth predicates (those that do not appear often in the ground truth), the network does poorly. Even when allowed to propose 100 tuples, the network does not offer the predicate. Figure 4 simply observes the fact that certain sets of relationship predicates appear predominantly in a subset of slots. No general explanation has been offered for this behaviour.<br />
<br />
== Conclusion ==<br />
In conclusion, the paper offers a novel approach to that enables the extraction of image semantics while perpetually reasoning over the entire context of the image. Associative embeddings are used to connect object and predicate relationships, and parallel “slots” allow for multiple detections in one pixel. While this approach offers noticeable improvements in accuracy, it is clear that work needs to be done to account for the non-uniform distributions of relationships in the dataset.<br />
<br />
== Appendices ==<br />
<br />
Appendix 1: Sample Outputs<br />
<br />
<br />
Appendix 2: Stacked Hourglass Architecture<br />
<br />
Although this goes beyond the focus of the paper, I would like to add a brief overview of the stacked hourglass architecture used to generate the heatmap. This architecture is unique in that it allows cyclical top-down, bottom-up inference and recombination of features. While most architectures focus on optimizing the bottom-up portion (reducing dimensionality), the stacked-hourglass gives the network more flexibility in how it generates a representation by allowing it to learn a series of down-sampling / up-sampling steps.<br />
<br />
== References ==<br />
1. Alejandro Newell and Jia Deng, “Pixels to Graphs by Associative Embedding,” in NIPS, 2017<br />
2. Alejandro Newell, Kaiyu Yang, and Jia Deng. Stacked Hourglass Networks for Human Pose Estimation. ECCV, 2016</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Pixels_to_Graphs_by_Associative_Embedding&diff=37166Pixels to Graphs by Associative Embedding2018-10-24T02:29:12Z<p>Gchalato: /* Introduction */</p>
<hr />
<div>== Introduction == <br />
The paper presents a novel approach to proposing a scene graph. A scene graph, as it relates to an image, is a graph with a vertex that represents each object identified in the image and an edge that represents relationships between the objects. <br />
<br />
An example of a scene graph:<br />
<br />
[[File:Scene Graph.PNG]]<br />
<br />
Current state-of-the-art techniques break down the construction of scene graphs by first identifying objects, and then predicting the edges for any given pair of identified objects. This paper introduces an architecture that defines the entire graph directly from the image, enabling the network to reason across the entirety of the image to understand relationships, as opposed to only predicting relationships using object labels. <br />
A key concern, given that the new architecture produces both vertices (objects) and edges (relationships) , is connecting the two. Specifically, the output of the network is some set of relationships E, and some set of Vertices V. The network needs to also output the “source” and “destination” of each relationship, so that the final graph can be formed. In the image above, for example, the network would also need to tell us that “holding” comes from “person” and goes to “Frisbee”. To do this, the paper uses associative embeddings. Specifically, the network outputs a particular “embedding vector” for each vertex, as well as a “source embedding” and “destination embedding” for each relationship. A final post-processing step finds the vertex embedding closest to each of the source / destination embeddings of each relationship and in this way assigns the edges to pairs of vertices.<br />
<br />
== The Architecture: == <br />
: '''1. Detecting Graph Elements'''<br />
<br />
Given an image of dimensions h x w, a stacked hourglass (Appendix 2) is used to generate a h x w x f representation of the image. It should be noted that the dimension of the output (which is non-trainable), to consider. Specifically, we need to have a resolution large enough to minimize the number of pixels with multiple detections while also being small enough to ensure that each 1 x 1 x f vector still contains the information needed for subsequent inference.<br />
A 1x1 convolution and sigmoid activation is performed on this result to generate a heatmap (one for objects and one for relationships, using separately determined convolutions). The value at a given pixel can be interpreted as the likelihood of detection at that particular pixel in the original image. <br />
In order to claim that there is an element at some pixel, we need to have some likelihood threshold. Then, if a given pixel in the map has a value >= the threshold, we claim that there is an element at that pixel. This threshold is calculated by using binary cross-entropy loss on the final values in the heatmap. Values with likelihoods greater than p-hat will be considered element detections. <br />
Finally, for each element that we detected, we extract the 1 x 1 x f feature vector. This is then used as an input to a set of FFNN, where we have a separate network for each characteristic of interest. The following image summarizes the process.<br />
<br />
:'''2. Connecting Elements with Associative Embeddings'''<br />
As explained earlier, to construct the scene graph, we need to know the source and destination of each edge. This is done through associative embeddings. <br />
First, let us define an embedding hi ϵ Rd produced for some vector i, and let us assume that we have n object detections in a particular image. Now, define hik, for k = 1 to Ki (where Ki is the number of edges in the graph with a vertex at vertex i) as the embedding associated with an edge that touches vertex i. We define two loss functions on these sets.<br />
<br />
The goal of Lpull is minimize the squared differences between the embedding of a given vertex and the embedding of an edge that references said vertex.<br />
<br />
On the other hand, minimizing Lpush implies assigning embeddings to vertices that are as far apart as possible. The further apart they are, the lower the output of max becomes, until eventually it reaches 0. Here, m is just a constant. In the paper, the values used were m = 8 and d = 8 (that is, 8D embeddings). Combining these two loss functions (and weighing them equally), accomplishes the task of predicting embeddings such that vertices are differentiated, but the embedding of a vertex is most similar to the vertex it references.<br />
<br />
:'''3. Support for Overlapping Detections'''<br />
An obvious concern is how the network would operate if there was more than one detection (be it object or relationship), in a given pixel. For example, detection of “shirt” and “person” may be centered at the exact same pixel. To account for this, the architecture is modified to allow for “slots” at each pixel. Specifically, so detections of objects are allowed at a particular pixel, while sr relationship detections are allowed at a given pixel. <br />
In order to allow for this, some changes are required after the feature extraction step. Specifically, we now use the 1x1xf vector as the input for so (or sr) different sets of 4 FFNNs, where the output is as shown in figure 2, but with the final FFNN outputting the probability of a detection existing in that particular slot, at that particular pixel. This new network is trained exclusively on whether or not a detection has been made in that slot, and, in prediction, is used to determine the number of slots to output at a given pixel. It is critical to note that this each of these so (or sr) sets of FFNNs share absolutely no weights. And each is trained for detection in its assigned slot.<br />
It is important to note that that this implies a change in the training procedure. We now have so (or sr) different predictions (be it class, or class + bounding box), that we need to match with our set of ground truth detections at a given pixel. Without this step, we would not be able to assign a value to the error for that sample. To do this, we match a one-hot encoded vector of the ground-truth class and bounding box anchor (the reference vector), and then match them with the so (or sr) outputs provided at a given pixel. The Hungarian method is used to ensure maximum matching between the outputs and the reference method while ensuring we do not assign the same detection to multiple slots. <br />
<br />
<br />
==Results==<br />
A quick note on notation: R@50 indicates what percentage of ground-truth subject-predicate-object tuples appeared in a proposal of 50 such tuples. Since R@100 offers more possibilities, it will necessarily be higher. The 6.7, for example, indicates that 6.7% of the ground truth pairs appeared in the proposals of the network. <br />
<br />
The authors tested the network against two other architectures designed to develop semantic understanding of images. For this, they used the Visual Genome dataset, with so = 3 and sr = 6. Overall, the new architecture vastly outperformed past models. The results were as follows:<br />
<br />
The table can be interpreted as follows:<br />
<br />
::'''SGGen (no RPN)''': Given a particular image, without the use of Region Proposal networks, the accuracy of the proposed scene graph. No class predictions are provided.<br />
::'''SGGen (with RPN)''': Same as above, except the output of the Region Proposal Network is used to enhance the input of a given image. No class predictions are provided.<br />
::'''SGCIs''': Ground-truth object bounding boxes are provided. The network is asked to classify them and determine relationships.<br />
::'''PredCIs''': As above, except the classes are also provided. The only goal is to predict relationships.<br />
<br />
Further analysis into the accuracy, when looking at predicates individually, shows that the architecture is very sensitive to over-represented relationship predicates.<br />
<br />
As shown in Figure 5, for many ground-truth predicates (those that do not appear often in the ground truth), the network does poorly. Even when allowed to propose 100 tuples, the network does not offer the predicate. Figure 4 simply observes the fact that certain sets of relationship predicates appear predominantly in a subset of slots. No general explanation has been offered for this behaviour.<br />
<br />
== Conclusion ==<br />
In conclusion, the paper offers a novel approach to that enables the extraction of image semantics while perpetually reasoning over the entire context of the image. Associative embeddings are used to connect object and predicate relationships, and parallel “slots” allow for multiple detections in one pixel. While this approach offers noticeable improvements in accuracy, it is clear that work needs to be done to account for the non-uniform distributions of relationships in the dataset.<br />
<br />
== Appendices ==<br />
<br />
Appendix 1: Sample Outputs<br />
<br />
<br />
Appendix 2: Stacked Hourglass Architecture<br />
<br />
Although this goes beyond the focus of the paper, I would like to add a brief overview of the stacked hourglass architecture used to generate the heatmap. This architecture is unique in that it allows cyclical top-down, bottom-up inference and recombination of features. While most architectures focus on optimizing the bottom-up portion (reducing dimensionality), the stacked-hourglass gives the network more flexibility in how it generates a representation by allowing it to learn a series of down-sampling / up-sampling steps.<br />
<br />
== References ==<br />
1. Alejandro Newell and Jia Deng, “Pixels to Graphs by Associative Embedding,” in NIPS, 2017<br />
2. Alejandro Newell, Kaiyu Yang, and Jia Deng. Stacked Hourglass Networks for Human Pose Estimation. ECCV, 2016</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Pixels_to_Graphs_by_Associative_Embedding&diff=37165Pixels to Graphs by Associative Embedding2018-10-24T02:29:02Z<p>Gchalato: /* Introduction */</p>
<hr />
<div>== Introduction == <br />
The paper presents a novel approach to proposing a scene graph. A scene graph, as it relates to an image, is a graph with a vertex that represents each object identified in the image and an edge that represents relationships between the objects. <br />
<br />
An example of a scene graph:<br />
[[File:Scene Graph.PNG]]<br />
<br />
Current state-of-the-art techniques break down the construction of scene graphs by first identifying objects, and then predicting the edges for any given pair of identified objects. This paper introduces an architecture that defines the entire graph directly from the image, enabling the network to reason across the entirety of the image to understand relationships, as opposed to only predicting relationships using object labels. <br />
A key concern, given that the new architecture produces both vertices (objects) and edges (relationships) , is connecting the two. Specifically, the output of the network is some set of relationships E, and some set of Vertices V. The network needs to also output the “source” and “destination” of each relationship, so that the final graph can be formed. In the image above, for example, the network would also need to tell us that “holding” comes from “person” and goes to “Frisbee”. To do this, the paper uses associative embeddings. Specifically, the network outputs a particular “embedding vector” for each vertex, as well as a “source embedding” and “destination embedding” for each relationship. A final post-processing step finds the vertex embedding closest to each of the source / destination embeddings of each relationship and in this way assigns the edges to pairs of vertices.<br />
<br />
== The Architecture: == <br />
: '''1. Detecting Graph Elements'''<br />
<br />
Given an image of dimensions h x w, a stacked hourglass (Appendix 2) is used to generate a h x w x f representation of the image. It should be noted that the dimension of the output (which is non-trainable), to consider. Specifically, we need to have a resolution large enough to minimize the number of pixels with multiple detections while also being small enough to ensure that each 1 x 1 x f vector still contains the information needed for subsequent inference.<br />
A 1x1 convolution and sigmoid activation is performed on this result to generate a heatmap (one for objects and one for relationships, using separately determined convolutions). The value at a given pixel can be interpreted as the likelihood of detection at that particular pixel in the original image. <br />
In order to claim that there is an element at some pixel, we need to have some likelihood threshold. Then, if a given pixel in the map has a value >= the threshold, we claim that there is an element at that pixel. This threshold is calculated by using binary cross-entropy loss on the final values in the heatmap. Values with likelihoods greater than p-hat will be considered element detections. <br />
Finally, for each element that we detected, we extract the 1 x 1 x f feature vector. This is then used as an input to a set of FFNN, where we have a separate network for each characteristic of interest. The following image summarizes the process.<br />
<br />
:'''2. Connecting Elements with Associative Embeddings'''<br />
As explained earlier, to construct the scene graph, we need to know the source and destination of each edge. This is done through associative embeddings. <br />
First, let us define an embedding hi ϵ Rd produced for some vector i, and let us assume that we have n object detections in a particular image. Now, define hik, for k = 1 to Ki (where Ki is the number of edges in the graph with a vertex at vertex i) as the embedding associated with an edge that touches vertex i. We define two loss functions on these sets.<br />
<br />
The goal of Lpull is minimize the squared differences between the embedding of a given vertex and the embedding of an edge that references said vertex.<br />
<br />
On the other hand, minimizing Lpush implies assigning embeddings to vertices that are as far apart as possible. The further apart they are, the lower the output of max becomes, until eventually it reaches 0. Here, m is just a constant. In the paper, the values used were m = 8 and d = 8 (that is, 8D embeddings). Combining these two loss functions (and weighing them equally), accomplishes the task of predicting embeddings such that vertices are differentiated, but the embedding of a vertex is most similar to the vertex it references.<br />
<br />
:'''3. Support for Overlapping Detections'''<br />
An obvious concern is how the network would operate if there was more than one detection (be it object or relationship), in a given pixel. For example, detection of “shirt” and “person” may be centered at the exact same pixel. To account for this, the architecture is modified to allow for “slots” at each pixel. Specifically, so detections of objects are allowed at a particular pixel, while sr relationship detections are allowed at a given pixel. <br />
In order to allow for this, some changes are required after the feature extraction step. Specifically, we now use the 1x1xf vector as the input for so (or sr) different sets of 4 FFNNs, where the output is as shown in figure 2, but with the final FFNN outputting the probability of a detection existing in that particular slot, at that particular pixel. This new network is trained exclusively on whether or not a detection has been made in that slot, and, in prediction, is used to determine the number of slots to output at a given pixel. It is critical to note that this each of these so (or sr) sets of FFNNs share absolutely no weights. And each is trained for detection in its assigned slot.<br />
It is important to note that that this implies a change in the training procedure. We now have so (or sr) different predictions (be it class, or class + bounding box), that we need to match with our set of ground truth detections at a given pixel. Without this step, we would not be able to assign a value to the error for that sample. To do this, we match a one-hot encoded vector of the ground-truth class and bounding box anchor (the reference vector), and then match them with the so (or sr) outputs provided at a given pixel. The Hungarian method is used to ensure maximum matching between the outputs and the reference method while ensuring we do not assign the same detection to multiple slots. <br />
<br />
<br />
==Results==<br />
A quick note on notation: R@50 indicates what percentage of ground-truth subject-predicate-object tuples appeared in a proposal of 50 such tuples. Since R@100 offers more possibilities, it will necessarily be higher. The 6.7, for example, indicates that 6.7% of the ground truth pairs appeared in the proposals of the network. <br />
<br />
The authors tested the network against two other architectures designed to develop semantic understanding of images. For this, they used the Visual Genome dataset, with so = 3 and sr = 6. Overall, the new architecture vastly outperformed past models. The results were as follows:<br />
<br />
The table can be interpreted as follows:<br />
<br />
::'''SGGen (no RPN)''': Given a particular image, without the use of Region Proposal networks, the accuracy of the proposed scene graph. No class predictions are provided.<br />
::'''SGGen (with RPN)''': Same as above, except the output of the Region Proposal Network is used to enhance the input of a given image. No class predictions are provided.<br />
::'''SGCIs''': Ground-truth object bounding boxes are provided. The network is asked to classify them and determine relationships.<br />
::'''PredCIs''': As above, except the classes are also provided. The only goal is to predict relationships.<br />
<br />
Further analysis into the accuracy, when looking at predicates individually, shows that the architecture is very sensitive to over-represented relationship predicates.<br />
<br />
As shown in Figure 5, for many ground-truth predicates (those that do not appear often in the ground truth), the network does poorly. Even when allowed to propose 100 tuples, the network does not offer the predicate. Figure 4 simply observes the fact that certain sets of relationship predicates appear predominantly in a subset of slots. No general explanation has been offered for this behaviour.<br />
<br />
== Conclusion ==<br />
In conclusion, the paper offers a novel approach to that enables the extraction of image semantics while perpetually reasoning over the entire context of the image. Associative embeddings are used to connect object and predicate relationships, and parallel “slots” allow for multiple detections in one pixel. While this approach offers noticeable improvements in accuracy, it is clear that work needs to be done to account for the non-uniform distributions of relationships in the dataset.<br />
<br />
== Appendices ==<br />
<br />
Appendix 1: Sample Outputs<br />
<br />
<br />
Appendix 2: Stacked Hourglass Architecture<br />
<br />
Although this goes beyond the focus of the paper, I would like to add a brief overview of the stacked hourglass architecture used to generate the heatmap. This architecture is unique in that it allows cyclical top-down, bottom-up inference and recombination of features. While most architectures focus on optimizing the bottom-up portion (reducing dimensionality), the stacked-hourglass gives the network more flexibility in how it generates a representation by allowing it to learn a series of down-sampling / up-sampling steps.<br />
<br />
== References ==<br />
1. Alejandro Newell and Jia Deng, “Pixels to Graphs by Associative Embedding,” in NIPS, 2017<br />
2. Alejandro Newell, Kaiyu Yang, and Jia Deng. Stacked Hourglass Networks for Human Pose Estimation. ECCV, 2016</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=File:Scene_Graph.PNG&diff=37164File:Scene Graph.PNG2018-10-24T02:27:05Z<p>Gchalato: </p>
<hr />
<div></div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=File:Sample_Pixel_Graph_Outputs.PNG&diff=37163File:Sample Pixel Graph Outputs.PNG2018-10-24T02:26:51Z<p>Gchalato: </p>
<hr />
<div></div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=File:Results_Table.PNG&diff=37162File:Results Table.PNG2018-10-24T02:26:26Z<p>Gchalato: </p>
<hr />
<div></div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=File:Results_-_Part_2.PNG&diff=37161File:Results - Part 2.PNG2018-10-24T02:25:57Z<p>Gchalato: </p>
<hr />
<div></div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=File:Loss_2.PNG&diff=37160File:Loss 2.PNG2018-10-24T02:25:45Z<p>Gchalato: </p>
<hr />
<div></div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=File:Loss_1.PNG&diff=37159File:Loss 1.PNG2018-10-24T02:25:36Z<p>Gchalato: </p>
<hr />
<div></div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=File:Hourglass.PNG&diff=37158File:Hourglass.PNG2018-10-24T02:25:20Z<p>Gchalato: </p>
<hr />
<div></div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=File:Extraction_Process.PNG&diff=37157File:Extraction Process.PNG2018-10-24T02:25:03Z<p>Gchalato: </p>
<hr />
<div></div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Pixels_to_Graphs_by_Associative_Embedding&diff=37155Pixels to Graphs by Associative Embedding2018-10-24T02:23:08Z<p>Gchalato: /* Results */</p>
<hr />
<div>== Introduction == <br />
The paper presents a novel approach to proposing a scene graph. A scene graph, as it relates to an image, is a graph with a vertex that represents each object identified in the image and an edge that represents relationships between the objects. <br />
<br />
<br />
Current state-of-the-art techniques break down the construction of scene graphs by first identifying objects, and then predicting the edges for any given pair of identified objects. This paper introduces an architecture that defines the entire graph directly from the image, enabling the network to reason across the entirety of the image to understand relationships, as opposed to only predicting relationships using object labels. <br />
A key concern, given that the new architecture produces both vertices (objects) and edges (relationships) , is connecting the two. Specifically, the output of the network is some set of relationships E, and some set of Vertices V. The network needs to also output the “source” and “destination” of each relationship, so that the final graph can be formed. In the image above, for example, the network would also need to tell us that “holding” comes from “person” and goes to “Frisbee”. To do this, the paper uses associative embeddings. Specifically, the network outputs a particular “embedding vector” for each vertex, as well as a “source embedding” and “destination embedding” for each relationship. A final post-processing step finds the vertex embedding closest to each of the source / destination embeddings of each relationship and in this way assigns the edges to pairs of vertices.<br />
<br />
== The Architecture: == <br />
: '''1. Detecting Graph Elements'''<br />
<br />
Given an image of dimensions h x w, a stacked hourglass (Appendix 2) is used to generate a h x w x f representation of the image. It should be noted that the dimension of the output (which is non-trainable), to consider. Specifically, we need to have a resolution large enough to minimize the number of pixels with multiple detections while also being small enough to ensure that each 1 x 1 x f vector still contains the information needed for subsequent inference.<br />
A 1x1 convolution and sigmoid activation is performed on this result to generate a heatmap (one for objects and one for relationships, using separately determined convolutions). The value at a given pixel can be interpreted as the likelihood of detection at that particular pixel in the original image. <br />
In order to claim that there is an element at some pixel, we need to have some likelihood threshold. Then, if a given pixel in the map has a value >= the threshold, we claim that there is an element at that pixel. This threshold is calculated by using binary cross-entropy loss on the final values in the heatmap. Values with likelihoods greater than p-hat will be considered element detections. <br />
Finally, for each element that we detected, we extract the 1 x 1 x f feature vector. This is then used as an input to a set of FFNN, where we have a separate network for each characteristic of interest. The following image summarizes the process.<br />
<br />
:'''2. Connecting Elements with Associative Embeddings'''<br />
As explained earlier, to construct the scene graph, we need to know the source and destination of each edge. This is done through associative embeddings. <br />
First, let us define an embedding hi ϵ Rd produced for some vector i, and let us assume that we have n object detections in a particular image. Now, define hik, for k = 1 to Ki (where Ki is the number of edges in the graph with a vertex at vertex i) as the embedding associated with an edge that touches vertex i. We define two loss functions on these sets.<br />
<br />
The goal of Lpull is minimize the squared differences between the embedding of a given vertex and the embedding of an edge that references said vertex.<br />
<br />
On the other hand, minimizing Lpush implies assigning embeddings to vertices that are as far apart as possible. The further apart they are, the lower the output of max becomes, until eventually it reaches 0. Here, m is just a constant. In the paper, the values used were m = 8 and d = 8 (that is, 8D embeddings). Combining these two loss functions (and weighing them equally), accomplishes the task of predicting embeddings such that vertices are differentiated, but the embedding of a vertex is most similar to the vertex it references.<br />
<br />
:'''3. Support for Overlapping Detections'''<br />
An obvious concern is how the network would operate if there was more than one detection (be it object or relationship), in a given pixel. For example, detection of “shirt” and “person” may be centered at the exact same pixel. To account for this, the architecture is modified to allow for “slots” at each pixel. Specifically, so detections of objects are allowed at a particular pixel, while sr relationship detections are allowed at a given pixel. <br />
In order to allow for this, some changes are required after the feature extraction step. Specifically, we now use the 1x1xf vector as the input for so (or sr) different sets of 4 FFNNs, where the output is as shown in figure 2, but with the final FFNN outputting the probability of a detection existing in that particular slot, at that particular pixel. This new network is trained exclusively on whether or not a detection has been made in that slot, and, in prediction, is used to determine the number of slots to output at a given pixel. It is critical to note that this each of these so (or sr) sets of FFNNs share absolutely no weights. And each is trained for detection in its assigned slot.<br />
It is important to note that that this implies a change in the training procedure. We now have so (or sr) different predictions (be it class, or class + bounding box), that we need to match with our set of ground truth detections at a given pixel. Without this step, we would not be able to assign a value to the error for that sample. To do this, we match a one-hot encoded vector of the ground-truth class and bounding box anchor (the reference vector), and then match them with the so (or sr) outputs provided at a given pixel. The Hungarian method is used to ensure maximum matching between the outputs and the reference method while ensuring we do not assign the same detection to multiple slots. <br />
<br />
<br />
==Results==<br />
A quick note on notation: R@50 indicates what percentage of ground-truth subject-predicate-object tuples appeared in a proposal of 50 such tuples. Since R@100 offers more possibilities, it will necessarily be higher. The 6.7, for example, indicates that 6.7% of the ground truth pairs appeared in the proposals of the network. <br />
<br />
The authors tested the network against two other architectures designed to develop semantic understanding of images. For this, they used the Visual Genome dataset, with so = 3 and sr = 6. Overall, the new architecture vastly outperformed past models. The results were as follows:<br />
<br />
The table can be interpreted as follows:<br />
<br />
::'''SGGen (no RPN)''': Given a particular image, without the use of Region Proposal networks, the accuracy of the proposed scene graph. No class predictions are provided.<br />
::'''SGGen (with RPN)''': Same as above, except the output of the Region Proposal Network is used to enhance the input of a given image. No class predictions are provided.<br />
::'''SGCIs''': Ground-truth object bounding boxes are provided. The network is asked to classify them and determine relationships.<br />
::'''PredCIs''': As above, except the classes are also provided. The only goal is to predict relationships.<br />
<br />
Further analysis into the accuracy, when looking at predicates individually, shows that the architecture is very sensitive to over-represented relationship predicates.<br />
<br />
As shown in Figure 5, for many ground-truth predicates (those that do not appear often in the ground truth), the network does poorly. Even when allowed to propose 100 tuples, the network does not offer the predicate. Figure 4 simply observes the fact that certain sets of relationship predicates appear predominantly in a subset of slots. No general explanation has been offered for this behaviour.<br />
<br />
== Conclusion ==<br />
In conclusion, the paper offers a novel approach to that enables the extraction of image semantics while perpetually reasoning over the entire context of the image. Associative embeddings are used to connect object and predicate relationships, and parallel “slots” allow for multiple detections in one pixel. While this approach offers noticeable improvements in accuracy, it is clear that work needs to be done to account for the non-uniform distributions of relationships in the dataset.<br />
<br />
== Appendices ==<br />
<br />
Appendix 1: Sample Outputs<br />
<br />
<br />
Appendix 2: Stacked Hourglass Architecture<br />
<br />
Although this goes beyond the focus of the paper, I would like to add a brief overview of the stacked hourglass architecture used to generate the heatmap. This architecture is unique in that it allows cyclical top-down, bottom-up inference and recombination of features. While most architectures focus on optimizing the bottom-up portion (reducing dimensionality), the stacked-hourglass gives the network more flexibility in how it generates a representation by allowing it to learn a series of down-sampling / up-sampling steps.<br />
<br />
== References ==<br />
1. Alejandro Newell and Jia Deng, “Pixels to Graphs by Associative Embedding,” in NIPS, 2017<br />
2. Alejandro Newell, Kaiyu Yang, and Jia Deng. Stacked Hourglass Networks for Human Pose Estimation. ECCV, 2016</div>Gchalatohttp://wiki.math.uwaterloo.ca/statwiki/index.php?title=Pixels_to_Graphs_by_Associative_Embedding&diff=37154Pixels to Graphs by Associative Embedding2018-10-24T02:18:17Z<p>Gchalato: </p>
<hr />
<div>== Introduction == <br />
The paper presents a novel approach to proposing a scene graph. A scene graph, as it relates to an image, is a graph with a vertex that represents each object identified in the image and an edge that represents relationships between the objects. <br />
<br />
<br />
Current state-of-the-art techniques break down the construction of scene graphs by first identifying objects, and then predicting the edges for any given pair of identified objects. This paper introduces an architecture that defines the entire graph directly from the image, enabling the network to reason across the entirety of the image to understand relationships, as opposed to only predicting relationships using object labels. <br />
A key concern, given that the new architecture produces both vertices (objects) and edges (relationships) , is connecting the two. Specifically, the output of the network is some set of relationships E, and some set of Vertices V. The network needs to also output the “source” and “destination” of each relationship, so that the final graph can be formed. In the image above, for example, the network would also need to tell us that “holding” comes from “person” and goes to “Frisbee”. To do this, the paper uses associative embeddings. Specifically, the network outputs a particular “embedding vector” for each vertex, as well as a “source embedding” and “destination embedding” for each relationship. A final post-processing step finds the vertex embedding closest to each of the source / destination embeddings of each relationship and in this way assigns the edges to pairs of vertices.<br />
<br />
== The Architecture: == <br />
: '''1. Detecting Graph Elements'''<br />
<br />
Given an image of dimensions h x w, a stacked hourglass (Appendix 2) is used to generate a h x w x f representation of the image. It should be noted that the dimension of the output (which is non-trainable), to consider. Specifically, we need to have a resolution large enough to minimize the number of pixels with multiple detections while also being small enough to ensure that each 1 x 1 x f vector still contains the information needed for subsequent inference.<br />
A 1x1 convolution and sigmoid activation is performed on this result to generate a heatmap (one for objects and one for relationships, using separately determined convolutions). The value at a given pixel can be interpreted as the likelihood of detection at that particular pixel in the original image. <br />
In order to claim that there is an element at some pixel, we need to have some likelihood threshold. Then, if a given pixel in the map has a value >= the threshold, we claim that there is an element at that pixel. This threshold is calculated by using binary cross-entropy loss on the final values in the heatmap. Values with likelihoods greater than p-hat will be considered element detections. <br />
Finally, for each element that we detected, we extract the 1 x 1 x f feature vector. This is then used as an input to a set of FFNN, where we have a separate network for each characteristic of interest. The following image summarizes the process.<br />
<br />
:'''2. Connecting Elements with Associative Embeddings'''<br />
As explained earlier, to construct the scene graph, we need to know the source and destination of each edge. This is done through associative embeddings. <br />
First, let us define an embedding hi ϵ Rd produced for some vector i, and let us assume that we have n object detections in a particular image. Now, define hik, for k = 1 to Ki (where Ki is the number of edges in the graph with a vertex at vertex i) as the embedding associated with an edge that touches vertex i. We define two loss functions on these sets.<br />
<br />
The goal of Lpull is minimize the squared differences between the embedding of a given vertex and the embedding of an edge that references said vertex.<br />
<br />
On the other hand, minimizing Lpush implies assigning embeddings to vertices that are as far apart as possible. The further apart they are, the lower the output of max becomes, until eventually it reaches 0. Here, m is just a constant. In the paper, the values used were m = 8 and d = 8 (that is, 8D embeddings). Combining these two loss functions (and weighing them equally), accomplishes the task of predicting embeddings such that vertices are differentiated, but the embedding of a vertex is most similar to the vertex it references.<br />
<br />
:'''3. Support for Overlapping Detections'''<br />
An obvious concern is how the network would operate if there was more than one detection (be it object or relationship), in a given pixel. For example, detection of “shirt” and “person” may be centered at the exact same pixel. To account for this, the architecture is modified to allow for “slots” at each pixel. Specifically, so detections of objects are allowed at a particular pixel, while sr relationship detections are allowed at a given pixel. <br />
In order to allow for this, some changes are required after the feature extraction step. Specifically, we now use the 1x1xf vector as the input for so (or sr) different sets of 4 FFNNs, where the output is as shown in figure 2, but with the final FFNN outputting the probability of a detection existing in that particular slot, at that particular pixel. This new network is trained exclusively on whether or not a detection has been made in that slot, and, in prediction, is used to determine the number of slots to output at a given pixel. It is critical to note that this each of these so (or sr) sets of FFNNs share absolutely no weights. And each is trained for detection in its assigned slot.<br />
It is important to note that that this implies a change in the training procedure. We now have so (or sr) different predictions (be it class, or class + bounding box), that we need to match with our set of ground truth detections at a given pixel. Without this step, we would not be able to assign a value to the error for that sample. To do this, we match a one-hot encoded vector of the ground-truth class and bounding box anchor (the reference vector), and then match them with the so (or sr) outputs provided at a given pixel. The Hungarian method is used to ensure maximum matching between the outputs and the reference method while ensuring we do not assign the same detection to multiple slots. <br />
<br />
<br />
==Results==<br />
A quick note on notation: R@50 indicates what percentage of ground-truth subject-predicate-object tuples appeared in a proposal of 50 such tuples. Since R@100 offers more possibilities, it will necessarily be higher. The 6.7, for example, indicates that 6.7% of the ground truth pairs appeared in the proposals of the network. <br />
<br />
The authors tested the network against two other architectures designed to develop semantic understanding of images. For this, they used the Visual Genome dataset, with so = 3 and sr = 6. Overall, the new architecture vastly outperformed past models. The results were as follows:<br />
<br />
The table can be interpreted as follows:<br />
<br />
**'''SGGen (no RPN)''': Given a particular image, without the use of Region Proposal networks, the accuracy of the proposed scene graph. No class predictions are provided.<br />
**'''SGGen (with RPN)''': Same as above, except the output of the Region Proposal Network is used to enhance the input of a given image. No class predictions are provided.<br />
**'''SGCIs''': Ground-truth object bounding boxes are provided. The network is asked to classify them and determine relationships.<br />
**'''PredCIs''': As above, except the classes are also provided. The only goal is to predict relationships.<br />
<br />
Further analysis into the accuracy, when looking at predicates individually, shows that the architecture is very sensitive to over-represented relationship predicates.<br />
<br />
As shown in Figure 5, for many ground-truth predicates (those that do not appear often in the ground truth), the network does poorly. Even when allowed to propose 100 tuples, the network does not offer the predicate. Figure 4 simply observes the fact that certain sets of relationship predicates appear predominantly in a subset of slots. No general explanation has been offered for this behaviour.<br />
<br />
== Conclusion ==<br />
In conclusion, the paper offers a novel approach to that enables the extraction of image semantics while perpetually reasoning over the entire context of the image. Associative embeddings are used to connect object and predicate relationships, and parallel “slots” allow for multiple detections in one pixel. While this approach offers noticeable improvements in accuracy, it is clear that work needs to be done to account for the non-uniform distributions of relationships in the dataset.<br />
<br />
== Appendices ==<br />
<br />
Appendix 1: Sample Outputs<br />
<br />
<br />
Appendix 2: Stacked Hourglass Architecture<br />
<br />
Although this goes beyond the focus of the paper, I would like to add a brief overview of the stacked hourglass architecture used to generate the heatmap. This architecture is unique in that it allows cyclical top-down, bottom-up inference and recombination of features. While most architectures focus on optimizing the bottom-up portion (reducing dimensionality), the stacked-hourglass gives the network more flexibility in how it generates a representation by allowing it to learn a series of down-sampling / up-sampling steps.<br />
<br />
== References ==<br />
1. Alejandro Newell and Jia Deng, “Pixels to Graphs by Associative Embedding,” in NIPS, 2017<br />
2. Alejandro Newell, Kaiyu Yang, and Jia Deng. Stacked Hourglass Networks for Human Pose Estimation. ECCV, 2016</div>Gchalato