Normalizing flow maps
We introduce Normalizing Flow Maps, exactly invertible generative transport maps that combine few-step sampling with tractable exact likelihoods, enabling scalable generation, likelihood-based post-training, and equilibrium sampling.