Difference between revisions of "conditional neural process"

From statwiki
Jump to: navigation, search
(Introduction)
Line 10: Line 10:
  
 
== Model ==
 
== Model ==
 +
Let training set be <math display="inline"> O =  \{x_i, y_i\}_{i = 0} ^ n-1</math>, and test set be  <math display="inline"> T =  \{x_i, y_i\}_{i = n} ^ {n + m - 1}</math>.
 +
 +
We assume the outputs are obtained by the following steps :
 +
 +
P be a probability distribution over functions  <math display="inline"> F : X \ti Y</math>

Revision as of 22:42, 18 November 2018

Introduction

To train a model effectively, deep neural networks require large datasets. To mitigate this data efficiency problem, learning in two phases is one approach : the first phase learns the statistics of a generic domain without committing to a specific learning task; the second phase learns a function for a specific task, but does so using only a small number of data points by exploiting the domain-wide statistics already learned.

For example, consider a data set [math] \{x_i, y_i\} [/math] with evaluations [math]y_i = f(x_i) [/math] for some unknown function [math]f[/math]. Assume [math]g[/math] is an approximating function of f. The aim is yo minimize the loss between [math]f[/math] and [math]g[/math] on the entire space [math]X[/math]. In practice, the routine is evaluated on a finite set of observations.

In this work, they proposed a family of models that represent solutions to the supervised problem, and ab end-to-end training approach to learning them, that combine neural networks with features reminiscent if Gaussian Process. They call this family of models Conditional Neural Processes.


Model

Let training set be [math] O = \{x_i, y_i\}_{i = 0} ^ n-1[/math], and test set be [math] T = \{x_i, y_i\}_{i = n} ^ {n + m - 1}[/math].

We assume the outputs are obtained by the following steps :

P be a probability distribution over functions [math] F : X \ti Y[/math]