Generalization Gap in Amortized Inference

Mingtian Zhang, Peter Hayes, David Barber

Keywords: variational inference, amortized inference, vae, lossless compression

Abstract: The ability of likelihood-based probabilistic models to generalize to unseen data is central to many machine learning applications such as lossless compression. In this work, we study the generalization of a popular class of probabilistic model - the Variational Auto-Encoder (VAE). We discuss the two generalization gaps that affect VAEs and show that overfitting is usually dominated by amortized inference. Based on this observation, we propose a new training objective that improves the generalization of amortized inference. We demonstrate how our method can improve performance in the context of image modeling and lossless compression.

TL;DR: We study and improve the generalization gap in amortized inference.

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:2205.11640/code)