Inference Dissection in Variational Autoencoders


Nov 03, 2017 (modified: Nov 03, 2017) ICLR 2018 Conference Blind Submission readers: everyone Show Bibtex
  • Abstract: It can be difficult to diagnose whether a given variational autoencoder can benefit from more costly inference and if so, how to improve the inference. This paper introduces quantitative tests to answer these questions. We categorize the ways that inference can fail in VAEs in order to gain insight into how to improve it. These tests also illustrate how the choice of the approximate posterior influences the trained model.
  • 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