From Variational to Deterministic Autoencoders: Difference between revisions
| Line 9: | Line 9: | ||
== Motivation == | == Motivation == | ||
The authors point to several drawbacks currently associated with VAE's including: | |||
- over-regularisation induced by the KL divergence term within the objective (Tolstikhin et al., 2017) | |||
- posterior collapse conjunction with powerful decoders (van den Oord et al., 2017) | |||
- increased variance in gradients caused by approximating expectations through sampling | |||
These issues motivate the assessment of whether the variational framework adopted by VAE's is necessary for generative modelling and whether an alternative framework utlizing regularization of the decoder in place of random noise could produce superior generative results. | |||
The authors are motivated to explore the use of regularisation to produce a smooth latent space where similar inputs are mapped to similar latent representation and for small variations in latent variables their decoded reconstruction vary only slightly. | |||
== Model Architecture == | == Model Architecture == | ||
Revision as of 01:15, 31 October 2020
Presented by
Partha Ghosh, Mehdi S. M. Sajjadi, Antonio Vergari, Michael Black, Bernhard Scholkopf
Introduction
This paper presents an alternative framework for generative modeling that is deterministic. They suggest that sampling from a stochastic encoder within a VAE can be interpreted as injecting noise into the input of a deterministic decoder and propose a framework for a regularized deterministic autoencoder (RAE) to generate samples that are comparable or better than those produced by VAE's.
Previous Work
Motivation
The authors point to several drawbacks currently associated with VAE's including: - over-regularisation induced by the KL divergence term within the objective (Tolstikhin et al., 2017) - posterior collapse conjunction with powerful decoders (van den Oord et al., 2017) - increased variance in gradients caused by approximating expectations through sampling
These issues motivate the assessment of whether the variational framework adopted by VAE's is necessary for generative modelling and whether an alternative framework utlizing regularization of the decoder in place of random noise could produce superior generative results.
The authors are motivated to explore the use of regularisation to produce a smooth latent space where similar inputs are mapped to similar latent representation and for small variations in latent variables their decoded reconstruction vary only slightly.