Research Seminar with Darius Faroughy (Rutgers University)

From Flow Matching to Flows in Function Space

Flow matching provides a simple framework for generative modeling by learning a time-dependent vector field that transports a tractable source distribution to a target distribution. In this talk, I will introduce the basic ideas behind flow matching and explain how this construction extends from finite-dimensional data to probability measures over functions. I will discuss the role of Gaussian-process reference measures, functional vector fields, and neural operators, and illustrate these ideas through our recent work on Prior-Fitted Functional Flows for pharmacokinetics, where the goal is to infer distributions over continuous concentration–time trajectories from sparse and irregular observations.