Improving Variational Autoencoders for New Physics Detection at the LHC with Normalizing Flows

TL;DR

This paper enhances variational autoencoders with normalizing flows to improve anomaly detection for new physics at the LHC, demonstrating how design choices impact performance and proposing methods for better detection accuracy.

Contribution

It introduces the integration of normalizing flows into VAEs for LHC anomaly detection, showing how this improves detection capabilities over baseline models.

Findings

Normalizing flows in latent space improve detection accuracy.

Design choices significantly affect model performance.

The approach outperforms baseline VAEs on benchmark datasets.

Abstract

We investigate how to improve new physics detection strategies exploiting variational autoencoders and normalizing flows for anomaly detection at the Large Hadron Collider. As a working example, we consider the DarkMachines challenge dataset. We show how different design choices (e.g., event representations, anomaly score definitions, network architectures) affect the result on specific benchmark new physics models. Once a baseline is established, we discuss how to improve the anomaly detection accuracy by exploiting normalizing flow layers in the latent space of the variational autoencoder.