Learning Discrete Structured Representations by Adversarially Maximizing Mutual Information

TL;DR

This paper introduces an adversarial method to learn discrete structured representations by maximizing mutual information, overcoming intractability issues, and demonstrating superior performance in document hashing tasks.

Contribution

It presents a novel adversarial approach to estimate mutual information for discrete representations, with a practical implementation using Markov distributions and binary encodings.

Findings

Outperforms existing discrete autoencoder baselines in document hashing

Produces highly compressed, interpretable representations

Reveals critical insights on variational prior choices

Abstract

We propose learning discrete structured representations from unlabeled data by maximizing the mutual information between a structured latent variable and a target variable. Calculating mutual information is intractable in this setting. Our key technical contribution is an adversarial objective that can be used to tractably estimate mutual information assuming only the feasibility of cross entropy calculation. We develop a concrete realization of this general formulation with Markov distributions over binary encodings. We report critical and unexpected findings on practical aspects of the objective such as the choice of variational priors. We apply our model on document hashing and show that it outperforms current best baselines based on discrete and vector quantized variational autoencoders. It also yields highly compressed interpretable representations.