Bayesian structure learning with generative flow networks

In this work, we propose to use a Generative Flow Networks (Gflownet, a novel class of probabilistic models) as an alternative to MCMC for approximating the posterior distribution over the structure of Bayesian networks, given a dataset of observations.