Perturb-and-max-product: Sampling and learning in discrete energy-based models This repo contains code for reproducing the results in the paper Pertur