Investigation Into the Viability of Neural Networks as a Means for Anomaly Detection in Experiments Like Atlas at the LHC

TL;DR

This paper explores the potential of neural networks to improve anomaly detection in large-scale particle physics experiments like Atlas at the LHC, aiming to identify rare events efficiently.

Contribution

It evaluates various neural network architectures for anomaly detection using simulated data, assessing their effectiveness in a high-energy physics context.

Findings

Neural networks show promise in detecting anomalies in simulated LHC data.

Certain architectures outperform traditional methods in identifying rare events.

The study provides insights into neural network suitability for real-world physics experiments.

Abstract

Petabytes of data are generated at the Atlas experiment at the Large Hadron Collider however not all of it is necessarily interesting, so what do we do with all of this data and how do we find these interesting needles in an uninteresting haystack. This problem can possibly be solved through the process of anomaly detection. In this document, Investigation Into the Viability of Neural Networks as a Means for Anomaly Detection in Experiments Like Atlas at the LHC the effectiveness of different types of neural network architectures as anomaly detectors are researched using Monte Carlo simulated data generated by the DarkMachines project. This data is meant to replicate Standard Model and Beyond Standard Model events. By finding an effective model, the Atlas experiment can become more effective and fewer interesting events will be lost.