Learning Discrete Structured Representations by Adversarially Maximizing Mutual Information
This addresses the problem of learning interpretable discrete representations for tasks like document hashing, offering a novel method for a known bottleneck in mutual information estimation.
The paper tackles learning discrete structured representations from unlabeled data by maximizing mutual information, using an adversarial objective to make this tractable. It shows improved performance over baselines in document hashing, yielding highly compressed and interpretable representations.
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.