Remasking Discrete Diffusion Models with Inference-Time Scaling

Published: 06 Mar 2025, Last Modified: 14 Apr 2025ICLR 2025 DeLTa Workshop PosterEveryoneRevisionsBibTeXCC BY 4.0
Track: long paper (up to 8 pages)
Keywords: discrete diffusion, deep generative models
TL;DR: Absorbing state discrete diffusion models that enable re-masking benefit from inference-time scaling, improving sample quality and controlled generation.
Abstract: Part of the success of diffusion models stems from their ability to perform iterative refinement, i.e., repeatedly correcting outputs during generation. However, modern masked discrete diffusion lacks this capability: when a token is generated, it cannot be updated again, even when it introduces an error. Here, we address this limitation by introducing the remasking diffusion model (ReMDM) sampler, a method that can be applied to pretrained masked diffusion models in a principled way and that is derived from a discrete diffusion model with a custom remasking backward process. Most interestingly, ReMDM endows discrete diffusion with a form of inference-time scaling. By increasing the number of sampling steps, ReMDM generates natural language outputs that approach the quality of autoregressive models, whereas when the computation budget is limited, ReMDM better maintains quality. ReMDM also improves sample quality of masked diffusion models for discretized images, and in scientific domains such as molecule design, ReMDM facilitates diffusion guidance and pushes the Pareto frontier of controllability relative to classical masking and uniform noise diffusion.
Submission Number: 43
Loading