Deep Dive into Semi-Supervised ELBO for Improving Classification Performance

TL;DR

This paper analyzes the ELBO decomposition in semi-supervised VAEs, identifies a mutual information decrease issue, and proposes a method to improve classification performance without losing generative capabilities.

Contribution

It introduces a novel approach to address mutual information loss in semi-supervised VAEs and enforces cluster assumptions to enhance classification accuracy.

Findings

Improved classification performance on diverse datasets.

Maintains generative power while enhancing classification.

Addresses mutual information decrease during ELBO optimization.

Abstract

Decomposition of the evidence lower bound (ELBO) objective of VAE used for density estimation revealed the deficiency of VAE for representation learning and suggested ways to improve the model. In this paper, we investigate whether we can get similar insights by decomposing the ELBO for semi-supervised classification using VAE model. Specifically, we show that mutual information between input and class labels decreases during maximization of ELBO objective. We propose a method to address this issue. We also enforce cluster assumption to aid in classification. Experiments on a diverse datasets verify that our method can be used to improve the classification performance of existing VAE based semi-supervised models. Experiments also show that, this can be achieved without sacrificing the generative power of the model.