Exploring Optimal Transport for Event-Level Anomaly Detection at the Large Hadron Collider

2024/01/28 by Nathaniel Craig, Craig, Nathaniel, J. Howard +3 · 3 citations
Physics and Astronomy · #Particle physics theoretical and experimental studies #Particle Detector Development and Performance #Radiation Detection and Scintillator Technologies

paper · pdf · doi:10.48550/arxiv.2401.15542

Abstract

Anomaly detection is a promising, model-agnostic strategy to find physics beyond the Standard Model. State-of-the-art machine learning methods offer impressive performance on anomaly detection tasks, but interpretability, resource, and memory concerns motivate considering a wide range of alternatives. We explore using the 2-Wasserstein distance from optimal transport theory, both as an anomaly score and as input to interpretable machine learning methods, for event-level anomaly detection at the Large Hadron Collider. The choice of ground space plays a key role in optimizing performance. We comment on the feasibility of implementing these methods in the L1 trigger system.

Cited by

• Optimal Transport Event Representation for Anomaly Detection
• Rare anomalies require large datasets: About proving the existence of anomalies
• Optimal Transport for e/π⁰ Particle Classification in LArTPC Neutrino Experiments

Related

• Optimal Transport Event Representation for Anomaly Detection
• New Physics Searches at the LHC through Event-based Anomaly Detection and Development of ADFilter Web-tool
• Anomaly Detection in Collider Physics via Factorized Observables
• Universal Anomaly Detection at the LHC: Transforming Optimal Classifiers and the DDD Method
• Efficiently Moving Instead of Reweighting Collider Events with Machine Learning