Discrete Latent Structure in Neural Networks

TL;DR

This paper reviews methods for learning discrete latent structures in neural networks, highlighting their common foundations and differences, to improve interpretability and structural bias in models across various data types.

Contribution

It unifies diverse strategies for discrete latent structure learning, revealing their fundamental similarities and differences, and provides insights into their applicability and properties.

Findings

Most methods share common building blocks

Connections between different learning strategies are clarified

Implications for model interpretability and structural bias

Abstract

Many types of data from fields including natural language processing, computer vision, and bioinformatics, are well represented by discrete, compositional structures such as trees, sequences, or matchings. Latent structure models are a powerful tool for learning to extract such representations, offering a way to incorporate structural bias, discover insight about the data, and interpret decisions. However, effective training is challenging, as neural networks are typically designed for continuous computation. This text explores three broad strategies for learning with discrete latent structure: continuous relaxation, surrogate gradients, and probabilistic estimation. Our presentation relies on consistent notations for a wide range of models. As such, we reveal many new connections between latent structure learning strategies, showing how most consist of the same small set of fundamental building blocks, but use them differently, leading to substantially different applicability and properties.