Machine Learning for Anomaly Detection in Particle Physics

TL;DR

This review surveys ML-driven anomaly detection in high-energy physics, emphasizing model-agnostic searches for new physics and real-time triggering. It categorizes training paradigms (supervised, weakly/semi-supervised, self-supervised, unsupervised) and details both overdensity and outlier approaches, including CWola, CATHODE, CURTAINs, ANODE, and NPLM. It also covers practical deployments in real-time triggers and detector monitoring, as well as emerging quantum-machine-learning perspectives with QSVMs, QNNs, and QAEs, highlighting current limitations and the need for rigorous validation. The discussion underscores challenges in validation, robustness to detector effects, and hardware constraints, while outlining the potential impact of scalable, model-agnostic anomaly detection in the HL-LHC era and beyond.

Abstract

The detection of out-of-distribution data points is a common task in particle physics. It is used for monitoring complex particle detectors or for identifying rare and unexpected events that may be indicative of new phenomena or physics beyond the Standard Model. Recent advances in Machine Learning for anomaly detection have encouraged the utilization of such techniques on particle physics problems. This review article provides an overview of the state-of-the-art techniques for anomaly detection in particle physics using machine learning. We discuss the challenges associated with anomaly detection in large and complex data sets, such as those produced by high-energy particle colliders, and highlight some of the successful applications of anomaly detection in particle physics experiments.

Figures (9)

• Figure 1: Weakly supervised density estimation techniques like CWola, CATHODE and Tag'N'Train take advantage of the fact that the signal (blue) can be localized in a specific region of phase space (SR). One can then use data sidebands (SB) to either estimate the background distribution (red) in the signal region, or to train a weakly supervised classifier (CWoLa) between SR and SB (figure from Ref. Hallin:2021wme).
• Figure 2: 95% confidence level upper limits on the crossection for a wide range of different signal models using the CWola bumphunt method ATLAS:2020iwa.
• Figure 3: An auto-encoder is trained to encode the input to a lower dimensional embedded space, and then decode it again in order to reconstruct the original input (top). The difference between the input and the output can be used as an anomaly score (bottom).
• Figure 4: The output, or anomaly score, of an auto-encoder trained on data collected by the ATLAS Experiment and evaluated on data (black) and a range of benchmark BSM signals ATLAS:2023ixc.
• Figure 5: Dijet mass distribution for the cluster with the highest signal-over-background ratio (right) and for the cluster closest to it (left). The background component in the signal region can be modeled from the closest cluster ucluster.