Normalizing Flow Maps

A generative modeling framework that connects any two time points along a path between a noise distribution and a data distribution through invertible mappings, generalizing normalizing flows. It enables both few-step sampling and exact likelihood computation.