Understanding the Mechanics of SPIGOT: Surrogate Gradients for Latent Structure Learning

TL;DR

This paper investigates surrogate gradient methods, particularly SPIGOT, for training latent structure models in language processing, providing a new perspective, algorithms, and empirical insights into their effectiveness and failure modes.

Contribution

It offers a principled motivation for SPIGOT and related estimators, introduces new algorithms, and compares their performance with existing methods.

Findings

SPIGOT and STE can be derived from a pulled-back objective perspective.

New algorithms in the same family outperform some existing estimators.

Empirical results reveal failure cases and practical insights for structured latent models.

Abstract

Latent structure models are a powerful tool for modeling language data: they can mitigate the error propagation and annotation bottleneck in pipeline systems, while simultaneously uncovering linguistic insights about the data. One challenge with end-to-end training of these models is the argmax operation, which has null gradient. In this paper, we focus on surrogate gradients, a popular strategy to deal with this problem. We explore latent structure learning through the angle of pulling back the downstream learning objective. In this paradigm, we discover a principled motivation for both the straight-through estimator (STE) as well as the recently-proposed SPIGOT - a variant of STE for structured models. Our perspective leads to new algorithms in the same family. We empirically compare the known and the novel pulled-back estimators against the popular alternatives, yielding new insight for practitioners and revealing intriguing failure cases.