Structured Voronoi Sampling

TL;DR

This paper introduces Structured Voronoi Sampling (SVS), a principled gradient-based method using Hamiltonian Monte Carlo for sampling from language models, improving control and diversity in text generation.

Contribution

The paper develops SVS, a new Hamiltonian Monte Carlo-based algorithm for principled sampling from language models, addressing the lack of theoretical grounding in gradient-based text generation methods.

Findings

SVS produces samples closer to the reference distribution than alternatives.

SVS generates fluent, diverse, and controlled text more effectively.

Empirical results demonstrate SVS's superiority in controlled generation tasks.

Abstract

Gradient-based sampling algorithms have demonstrated their effectiveness in text generation, especially in the context of controlled text generation. However, there exists a lack of theoretically grounded and principled approaches for this task. In this paper, we take an important step toward building a principled approach for sampling from language models with gradient-based methods. We use discrete distributions given by language models to define densities and develop an algorithm based on Hamiltonian Monte Carlo to sample from them. We name our gradient-based technique Structured Voronoi Sampling (SVS). In an experimental setup where the reference distribution is known, we show that the empirical distribution of SVS samples is closer to the reference distribution compared to alternative sampling schemes. Furthermore, in a controlled generation task, SVS is able to generate fluent and diverse samples while following the control targets significantly better than other methods.