= Inference Suboptimality in Variational Autoencoders by Chris Cremer and others
{c}
{tag=Work by Chris Cremer}
{tag=Work by Xuechen Li}
{tag=Work by David Duvenaud}
{title2=2018}
https://proceedings.mlr.press/v80/cremer18a.html
Primary PDF: https://proceedings.mlr.press/v80/cremer18a/cremer18a.pdf . Title-page byline: <Chris Cremer>, <Xuechen Li> and <David Duvenaud>; PMLR80,1078–1086(2018). Table1 on printed page1079 defines the inference, approximation and amortization gaps; the introduction specifies the optimal per-observation approximation within the chosen family.
https://proceedings.mlr.press/v80/cremer18a/cremer18a.pdf[§3.1 equation8, approximation/amortization decomposition; §3.2 and Figure2, expressive approximations; §3.4, separate optimization per observation and 100 Monte Carlo samples per bound estimate]
Back to article page
