Discouraging posterior collapse in hierarchical Variational Autoencoders using context

TL;DR

This paper investigates posterior collapse in hierarchical VAEs and proposes a novel approach using a top-level context with Discrete Cosine Transform to improve latent space utilization without sacrificing generative quality.

Contribution

It introduces a hierarchical VAE with a top-level context via Discrete Cosine Transform to mitigate posterior collapse, a problem often overlooked in hierarchical models.

Findings

Improved latent space utilization observed

No negative impact on generative performance

Effective reduction of posterior collapse

Abstract

Hierarchical Variational Autoencoders (VAEs) are among the most popular likelihood-based generative models. There is a consensus that the top-down hierarchical VAEs allow effective learning of deep latent structures and avoid problems like posterior collapse. Here, we show that this is not necessarily the case, and the problem of collapsing posteriors remains. To discourage this issue, we propose a deep hierarchical VAE with a context on top. Specifically, we use a Discrete Cosine Transform to obtain the last latent variable. In a series of experiments, we observe that the proposed modification allows us to achieve better utilization of the latent space and does not harm the model's generative abilities.