Inference Suboptimality in Variational Autoencoders

Chris Cremer, Xuechen Li, David Duvenaud

Abstract: Amortized inference has led to efficient approximate inference for large datasets. The quality of posterior inference is largely determined by two factors: a) the ability of the variational distribution to model the true posterior and b) the capacity of the recognition network to generalize inference over all datapoints. We analyze approximate inference in variational autoencoders in terms of these factors. We find that suboptimal inference is often due to amortizing inference rather than the limited complexity of the approximating distribution. We show that this is due partly to the generator learning to accommodate the choice of approximation. Furthermore, we show that the parameters used to increase the expressiveness of the approximation play a role in generalizing inference rather than simply improving the complexity of the approximation.

TL;DR: We decompose the gap between the marginal log-likelihood and the evidence lower bound and study the effect of the approximate posterior on the true posterior distribution in VAEs.

Keywords: Approximate Inference, Amortization, Posterior Approximations, Variational Autoencoder

Community Implementations: [ 2 code implementations](https://www.catalyzex.com/paper/arxiv:1801.03558/code)