Bayesian Modelling and Monte Carlo Inference for GAN

Bayesian modelling is a principal framework to perform model aggregation, which has been a primary mechanism to combat mode collapsing in the context of Generative Adversarial Networks (GANs). In this paper, we propose a novel Bayesian modelling framework for GANs, which iteratively learns a distribution over generators. We tailor stochastic gradient Hamiltonian Monte Carlo with novel gradient approximation to perform Bayesian inference. Theoretically, we prove any generator distribution which produces the target data distribution is an equilibrium of our algorithm. Empirical evidence on categorical distributed data and synthetic high-dimensional multi-modal data demonstrates the superior performance of our method over the start-of-art multigenerator and other Bayesian treatment for GANs.