Peter Pao-Huang introduces Flux Matching for generative modeling
Peter Pao-Huang introduced Flux Matching with collaborators including Stefano Ermon and Xiaojie Qiu. The paradigm extends diffusion models by supporting arbitrary vector fields instead of score functions and incorporates structural priors into the dynamics. Visual comparisons show Flux Matching producing spirals and loops versus radial bursts under score matching. Stefano Ermon highlighted its fast mixing properties. Kosta Derpanis of York University reposted the announcement.
Coolest thing in the world 🤓
Introducing Flux Matching, a generative modeling paradigm that generalizes diffusion models to vector fields that need not be the score function. Enables structural priors in the dynamics, faster sampling, interpretable generation, and more! w/ @StefanoErmon @Xiaojie_Qiu 🧵⤵️
Excited to see my student’s work on Flux Matching out. It turns out you can learn a much broader class of vector fields with the data distribution as stationary (not just the score). This lets you enforce useful properties like fast mixing, and it already works on high-dimensional image datasets!
Introducing Flux Matching, a generative modeling paradigm that generalizes diffusion models to vector fields that need not be the score function. Enables structural priors in the dynamics, faster sampling, interpretable generation, and more! w/ @StefanoErmon @Xiaojie_Qiu 🧵⤵️