Better Latent Spaces for Better Autoencoders

TL;DR

This paper introduces novel latent space configurations for autoencoders, enhancing their performance and interpretability in anomaly detection tasks at the LHC by deriving classifiers from these spaces.

Contribution

It proposes using Gaussian mixture and Dirichlet latent spaces in autoencoders, with the Dirichlet setup notably improving both performance and interpretability.

Findings

Dirichlet latent space improves autoencoder interpretability

Classifiers derived from latent spaces enhance anomaly detection

Gaussian mixture models provide flexible latent representations

Abstract

Autoencoders as tools behind anomaly searches at the LHC have the structural problem that they only work in one direction, extracting jets with higher complexity but not the other way around. To address this, we derive classifiers from the latent space of (variational) autoencoders, specifically in Gaussian mixture and Dirichlet latent spaces. In particular, the Dirichlet setup solves the problem and improves both the performance and the interpretability of the networks.