Deformations of Boltzmann Distributions

Abstract
Consider a one-parameter family of Boltzmann distributions . This work studies the problem of sampling from by first sampling from and then applying a transformation so that the transformed samples follow . We derive an equation relating and the corresponding family of unnormalized log-likelihoods . The utility of this idea is demonstrated on the lattice field theory by extending its defining action to a family of actions and finding a such that normalizing flows perform better at learning the Boltzmann distribution than at learning .
View on arXivComments on this paper