Conditional Flow Matching: Simulation-Free Dynamic Optimal Transport
Alex Tong, postdoctoral fellow at MILA (Université de Montréal), will talk about conditional flow matching, a family of simulation-free training objectives for continuous normalizing flows.

Abstract

Flow matching models learn a (possibly stochastic) mapping between source and target distributions. Common paradigms include diffusion models, score matching models, and continuous normalizing flows. In this talk I will discuss how incorporating ideas from optimal transport can lead to improved training and inference of flow matching models by reducing variance of the objective and creating “straighter” paths that can be integrated in fewer steps. I will then show how these methods can be used in the domains of image generation, modelling cell trajectories, and protein generation.

References

In this series →

Probabilistic Models
Trainings: An introduction to Bayesian methods in … Seminar: Predictive, Scalable and Interpretable … Pills: SoftFlow: Probabilistic Framework for … Seminar: Joint Probability Trees Pills: Flow Matching for Generative Modeling Pills: Random Sum-Product Networks: A Simple … Pills: Joint Probability Trees Pills: Spectral likelihood expansions for … Pills: Towards Generative Modelling of … Pills: Statistical Rethinking Pills: Natural Posterior Networks Pills: Variational Methods for Simulation Based … Pills: Scalable Bayesian Deep Learning with … Seminar: Hamiltonian Monte Carlo Sampling Seminar: Bayesian ML and Probabilistic … Seminar: Bayesian ML and Probabilistic … Seminar: Bayesian ML and Probabilistic …