Verlet Flows: Exact-Likelihood Integrators for Flow-Based Generative Models

arXiv:2405.02805v1 Announce Type: new
Abstract: Approximations in computing model likelihoods with continuous normalizing flows (CNFs) hinder the use of these models for importance sampling of Boltzmann distributions, where exact likelihoods are required. In this work, we present Verlet flows, a class of CNFs on an augmented state-space inspired by symplectic integrators from Hamiltonian dynamics. When used with carefully constructed Taylor-Verlet integrators, Verlet flows provide exact-likelihood generative models which generalize coupled flow architectures from a non-continuous setting while imposing minimal expressivity …

abstract arxiv boltzmann class computing continuous continuous normalizing flows cs.lg flow generative generative models hinder importance likelihood sampling space state type work