Training-free guidance of diffusion models for generalised inpainting
Keywords: generative, diffusion, sampling, guidance, langevin, mcmc, images, inpainting, proteins, t-cells
Abstract: Diffusion models facilitate powerful control over the generative process. Here we introduce training-free guidance, a method for sampling from a broad class of conditional distributions that can be considered generalisations of inpainting. The method is grounded in annealed Langevin dynamics which ensures convergence to the exact conditional distribution, unlike existing methods for inpainting which rely on heuristics. We demonstrate training-free guidance using pretrained unconditional models for image, protein structure, and protein sequence generation and improve upon state-of-the-art approaches. We show the versatility of training-free guidance by addressing a wide range of tasks, including multi-motif scaffolding and amino acid mutagenesis of T cell receptors.
Primary Area: generative models
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 11388
Loading