1. Loss Extended
With an overall view of VAE, we can extend standard VAE loss in the reconstruction loss and regularization loss, for the original VAE, which is the KL term, but there are all kinds of versions which we can treat them as different combination of regularization methods
2. The architecture of different regularizer
So we show how to decouple the standard KL loss into different terms, now we could analyze losses with distributions together and show how to decouple them and how they related to each other
3. Another Perspective
4. R-D trade-off
The rate-distortion-usefulness tradeoff. Here we argue that even if one is able to reach any desired rate-distortion tradeoff point, in particular targeting a representation with specific rate R, the learned representation might still be useless for a specific downstream task. This stems from the fact that
(i) it is unclear which part of the total information (entropy) is stored in z and which part is stored in the decoder, and
(ii) even if the information relevant for the downstream task is stored in z, there is no guarantee that it is stored in a form that can be exploited by the model used to solve the downstream task.
For example, regarding
(i), if the downstream task is an image classification task, the representation
should store the object class or the most prominent object features. On the other hand, if the down-stream task is to recognize the relative ordering of objects, the locations have to be encoded instead. Concerning (ii), if we use a linear model on top of the representation as often done in practice, the
representation needs to have structure amenable to linear prediction.
5. Overview
Reference
1. Recent Advances in Autoencoder-Based Representation Learning
2. https://github.com/YannDubs/disentangling-vae
3. VARIATIONAL INFERENCE OF DISENTANGLED LATENT CONCEPTS FROM UNLABELED OBSERVATIONS