generative flow networks (GFlowNets) are a method for learning a stochastic
policy for generating compositional objects, such as graphs or strings, from a
given unnormalized density by sequences of actions, where many possible action
sequences may lead to the same object. We find previ