Learning Hierarchical Features from Generative Models

TL;DR

This paper reveals limitations of current hierarchical generative models in learning hierarchical features and introduces an alternative architecture that effectively learns interpretable, disentangled features without task-specific priors.

Contribution

The paper proves existing variational methods do not leverage hierarchical structures and proposes a new architecture that overcomes these limitations.

Findings

Existing models do not utilize hierarchical structure effectively

The proposed model learns interpretable, disentangled features

Model performs well on natural image datasets

Abstract

Deep neural networks have been shown to be very successful at learning feature hierarchies in supervised learning tasks. Generative models, on the other hand, have benefited less from hierarchical models with multiple layers of latent variables. In this paper, we prove that hierarchical latent variable models do not take advantage of the hierarchical structure when trained with existing variational methods, and provide some limitations on the kind of features existing models can learn. Finally we propose an alternative architecture that do not suffer from these limitations. Our model is able to learn highly interpretable and disentangled hierarchical features on several natural image datasets with no task specific regularization or prior knowledge.