Continuous diffusion for categorical data

TL;DR

This paper introduces CDCD, a continuous diffusion framework for categorical data like language, maintaining the benefits of continuous models and demonstrating effectiveness across language tasks.

Contribution

The paper presents a novel continuous diffusion approach for categorical data, bridging the gap between continuous diffusion models and discrete data types.

Findings

Effective language modeling performance

Preserves continuous diffusion benefits for categorical data

Demonstrates versatility across multiple language tasks

Abstract

Diffusion models have quickly become the go-to paradigm for generative modelling of perceptual signals (such as images and sound) through iterative refinement. Their success hinges on the fact that the underlying physical phenomena are continuous. For inherently discrete and categorical data such as language, various diffusion-inspired alternatives have been proposed. However, the continuous nature of diffusion models conveys many benefits, and in this work we endeavour to preserve it. We propose CDCD, a framework for modelling categorical data with diffusion models that are continuous both in time and input space. We demonstrate its efficacy on several language modelling tasks.