B1600
Title: Posterior sampling and estimation of model evidence using neural networks
Authors: George Cantwell - University of Cambridge (United Kingdom) [presenting]
Abstract: The closely related problems of sampling from a distribution known up to a normalizing constant and estimating said normalizing are considered. It is shown how variational autoencoders (VAEs) can be applied to this task. In their standard applications, VAEs are trained to fit data drawn from an intractable distribution. The logic and training of the VAE can be inverted to fit a simple and tractable distribution on the assumption of a complex and intractable latent distribution specified up to normalization. This procedure constructs approximations without training data or Markov chain Monte Carlo sampling. The method is illustrated with three examples: the Ising model, graph clustering, and ranking.
