Kernel Implicit Variational Inference

Recent progress in variational inference has paid much attention to the flexibility of variational posteriors. Work has been done to use implicit distributions, i.e., distributions without tractable likelihoods as the variational posterior. However, existing methods on implicit posteriors still face challenges of noisy estimation and can hardly scale to high-dimensional latent variable models. In this paper, we present an implicit variational inference approach with kernel density ratio fitting that addresses these challenges. As far as we know, for the first time implicit variational inference is successfully applied to Bayesian neural networks, which shows promising results on both regression and classification tasks.