Exploring Diffusion and Flow Matching Under Generator Matching

TL;DR

This paper provides a unified theoretical comparison of diffusion and flow matching within the Generator Matching framework, explaining their differences, robustness, and potential for new model classes.

Contribution

It introduces a unified Markov framework to analyze diffusion and flow matching, revealing their relationships and guiding the construction of hybrid models.

Findings

Flow matching models are more robust empirically.

Unified framework explains differences between diffusion and flow matching.

Potential for creating new hybrid generative models.

Abstract

In this paper, we present a comprehensive theoretical comparison of diffusion and flow matching under the Generator Matching framework. Despite their apparent differences, both diffusion and flow matching can be viewed under the unified framework of Generator Matching. By recasting both diffusion and flow matching under the same generative Markov framework, we provide theoretical insights into why flow matching models can be more robust empirically and how novel model classes can be constructed by mixing deterministic and stochastic components. Our analysis offers a fresh perspective on the relationships between state-of-the-art generative modeling paradigms.