Disentangling Learning Representations with Density Estimation

TL;DR

This paper introduces GCAE, a novel autoencoder that reliably disentangles learned representations by using flexible density estimation and the DTC metric to manage high-dimensional latent spaces effectively.

Contribution

GCAE employs a new approach to disentanglement by decomposing the latent space into low-dimensional subsets with density estimation, improving reliability over existing methods.

Findings

GCAE achieves competitive disentanglement scores.

It demonstrates improved reliability in disentanglement tasks.

The method effectively manages high-dimensional latent spaces.

Abstract

Disentangled learning representations have promising utility in many applications, but they currently suffer from serious reliability issues. We present Gaussian Channel Autoencoder (GCAE), a method which achieves reliable disentanglement via flexible density estimation of the latent space. GCAE avoids the curse of dimensionality of density estimation by disentangling subsets of its latent space with the Dual Total Correlation (DTC) metric, thereby representing its high-dimensional latent joint distribution as a collection of many low-dimensional conditional distributions. In our experiments, GCAE achieves highly competitive and reliable disentanglement scores compared with state-of-the-art baselines.