Finding New Physics without learning about it: Anomaly Detection as a tool for Searches at Colliders

TL;DR

This paper introduces a novel anomaly detection approach for collider data to identify potential new physics phenomena without prior knowledge of their specific signatures, leveraging machine learning techniques trained on Standard Model events.

Contribution

It presents three new anomaly detection methods for high-energy physics and compares their effectiveness to supervised neural networks in discovering new physics signals.

Findings

Semi-supervised anomaly detection shows high sensitivity to new physics signals.

Deep and shallow AD methods outperform traditional supervised models in certain scenarios.

The approach is promising for analyzing current and future collider data.

Abstract

In this paper we propose a new strategy, based on anomaly detection methods, to search for new physics phenomena at colliders independently of the details of such new events. For this purpose, machine learning techniques are trained using Standard Model events, with the corresponding outputs being sensitive to physics beyond it. We explore three novel AD methods in HEP: Isolation Forest, Histogram Based Outlier Detection, and Deep Support Vector Data Description; alongside the most customary Autoencoder. In order to evaluate the sensitivity of the proposed approach, predictions from specific new physics models are considered and compared to those achieved when using fully supervised deep neural networks. A comparison between shallow and deep anomaly detection techniques is also presented. Our results demonstrate the potential of semi-supervised anomaly detection techniques to extensively explore the present and future hadron colliders' data.