Improving Function Space Flow Matching with Kernel Optimal Transport

License: See LICENSE file · Updated 2026-09-29
My rating

The paper addresses generative modeling of distributions in infinite-dimensional function spaces, with examples including time series and solutions of partial differential equations. It builds on Functional Flow Matching, which transports a Gaussian prior to the data distribution through a learned velocity field, and considers kernel optimal transport for improving the independent endpoint pairing used during training.

Tags: