Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
We introduce a novel neural algorithm for sampling from unnormalized Boltzmann densities by leveraging the Jordan--Kinderlehrer--Otto (JKO) scheme within the Wasserstein--Fisher--Rao (WFR) geometry. We establish theoretical convergence guarantees, proving exponential speed for WFR JKO iterates to the target distribution irrespective of structural assumptions like log-concavity. The practical implementation utilizes reweighted normalizing flows to parametrize the transport and reaction operators, demonstrating strong performance on multimodal targets.
Reweighted normalizing flows enable exponential convergence to arbitrary Boltzmann distributions via a novel WFR JKO scheme, bypassing common structural assumptions.
We propose a neural algorithm for sampling from distributions specified by unnormalized Boltzmann densities. Our approach is based on the Jordan--Kinderlehrer--Otto scheme for the Kullback--Leibler divergence in the Wasserstein--Fisher--Rao geometry (WFR JKO scheme). Our contributions are twofold. First, we prove that, for any fixed step size, the exact WFR JKO iterates converge exponentially fast to the target as the number of iterations tends to infinity. Notably, this result requires no structural assumptions on the target, such as log-concavity or a logarithmic Sobolev inequality. Second, we develop a neural implementation of the WFR JKO scheme that parametrizes its transport and reaction components using reweighted normalizing flows. Numerical experiments on challenging multimodal targets demonstrate the promising performance of the proposed method.