Our paper "On the Guidance of Flow Matching" is selected as a Spotlight Poster at ICML'25.

In this paper, we proposed a new theoretical framework for the energy guidance of flow matching, providing a new perspective to execute theoretically justified plug-and-play guidance for flow matching models.

Previously, the energy guidance of diffusion models can be more straightforwardly derived from score matching, which is not the case for flow matching models. Thus, we derived the exact energy guidance for flow matching models in a different way. Then, based on the new framework, we further propose a family of different approximation guidance methods, and evaluated their performance on different tasks including decision making and image inverse problems.

Please find more details of our work in the paper.

This year, ICML received 12,107 submissions (not including desk-rejected papers). From these submissions, 3,260 were accepted, giving an acceptance rate of 26.9\%. Only 313 were designated as “spotlight posters” – representing the top 2.6\% of all submissions. The spotlights represent the submissions most highly-recommended by the program committee. So it’s really encouraging to see our work being recognized by the community and I am thankful to my co-authors for their help and support.