On Posterior Collapse and Encoder Feature Dispersion in Sequence VAEs

TL;DR

This paper investigates the causes of posterior collapse in sequence VAEs for text modeling, identifying encoder feature dispersion as a key factor, and proposes a simple pooling method to prevent collapse and improve performance.

Contribution

It introduces a straightforward pooling technique that enhances encoder feature dispersion, effectively preventing posterior collapse in sequence VAEs for natural language processing.

Findings

Pooling prevents posterior collapse effectively.

The method achieves better data log-likelihood scores.

It is more computationally efficient than existing solutions.

Abstract

Variational autoencoders (VAEs) hold great potential for modelling text, as they could in theory separate high-level semantic and syntactic properties from local regularities of natural language. Practically, however, VAEs with autoregressive decoders often suffer from posterior collapse, a phenomenon where the model learns to ignore the latent variables, causing the sequence VAE to degenerate into a language model. In this paper, we argue that posterior collapse is in part caused by the lack of dispersion in encoder features. We provide empirical evidence to verify this hypothesis, and propose a straightforward fix using pooling. This simple technique effectively prevents posterior collapse, allowing model to achieve significantly better data log-likelihood than standard sequence VAEs. Comparing to existing work, our proposed method is able to achieve comparable or superior performances while being more computationally efficient.