All Posts

Published on
August 3, 2026
From Spread MLE to Implicit MLE
IMLE Spread-Divergence Maximum-Likelihood Implicit-Models
We show that IMLE can be viewed as the zero-noise hard-assignment limit of a finite-sample approximation to spread maximum likelihood.
Published on
October 1, 2023
Generalization Gaps of VAE
VAE Generalization
We discuss the two generalization gaps that can affect the test likelihood of VAEs and show that the overfitting of VAE is usually dominated by the overfitting of amortized inference.
Published on
December 31, 2021
A Failure Case of VAE
VAE Manifold
We discuss a simple but important failure mode of VAEs as well as s solution proposed in paper Spread Divergences to fix the broken VAE.

From Spread MLE to Implicit MLE