Improving Discrete Diffusion Models via Structured Preferential Generation

TL;DR

This paper introduces a structured forward process for discrete diffusion models that exploits the hierarchical nature of discrete data, leading to improved likelihood scores in text generation tasks.

Contribution

It presents a novel structured diffusion process that leverages data hierarchy, enhancing performance of discrete diffusion models over previous approaches.

Findings

Improved log-likelihood scores on text8 dataset

Biases in generative process enhance discrete data modeling

Potential for significant performance gains in discrete diffusion models

Abstract

In the domains of image and audio, diffusion models have shown impressive performance. However, their application to discrete data types, such as language, has often been suboptimal compared to autoregressive generative models. This paper tackles the challenge of improving discrete diffusion models by introducing a structured forward process that leverages the inherent information hierarchy in discrete categories, such as words in text. Our approach biases the generative process to produce certain categories before others, resulting in a notable improvement in log-likelihood scores on the text8 dataset. This work paves the way for more advances in discrete diffusion models with potentially significant enhancements in performance.