English

Amortizing intractable inference in diffusion models for vision, language, and control

Machine Learning 2025-01-14 v2 Computer Vision and Pattern Recognition

Abstract

Diffusion models have emerged as effective distribution estimators in vision, language, and reinforcement learning, but their use as priors in downstream tasks poses an intractable posterior inference problem. This paper studies amortized sampling of the posterior over data, xppost(x)p(x)r(x)\mathbf{x}\sim p^{\rm post}(\mathbf{x})\propto p(\mathbf{x})r(\mathbf{x}), in a model that consists of a diffusion generative model prior p(x)p(\mathbf{x}) and a black-box constraint or likelihood function r(x)r(\mathbf{x}). We state and prove the asymptotic correctness of a data-free learning objective, relative trajectory balance, for training a diffusion model that samples from this posterior, a problem that existing methods solve only approximately or in restricted cases. Relative trajectory balance arises from the generative flow network perspective on diffusion models, which allows the use of deep reinforcement learning techniques to improve mode coverage. Experiments illustrate the broad potential of unbiased inference of arbitrary posteriors under diffusion priors: in vision (classifier guidance), language (infilling under a discrete diffusion LLM), and multimodal data (text-to-image generation). Beyond generative modeling, we apply relative trajectory balance to the problem of continuous control with a score-based behavior prior, achieving state-of-the-art results on benchmarks in offline reinforcement learning.

Keywords

Cite

@article{arxiv.2405.20971,
  title  = {Amortizing intractable inference in diffusion models for vision, language, and control},
  author = {Siddarth Venkatraman and Moksh Jain and Luca Scimeca and Minsu Kim and Marcin Sendera and Mohsin Hasan and Luke Rowe and Sarthak Mittal and Pablo Lemos and Emmanuel Bengio and Alexandre Adam and Jarrid Rector-Brooks and Yoshua Bengio and Glen Berseth and Nikolay Malkin},
  journal= {arXiv preprint arXiv:2405.20971},
  year   = {2025}
}

Comments

NeurIPS 2024; code: https://github.com/GFNOrg/diffusion-finetuning

R2 v1 2026-06-28T16:48:40.032Z