Speaker
Description
In this study, we investigate the use of state-of-the-art classifiers, including graph networks and transformer-based architectures, for anomaly detection at the Large Hadron Collider (LHC). Traditionally, these classifiers have been used primarily for classification (signal or background) tasks. However, we investigate their potential as anomaly detection algorithms with the goal of identifying rare and unusual events in high-energy physics data.
We evaluate the performance of these classifiers converted to anomaly detectors on two data sets: 4-top signals and the Darkmachines anomaly detection data.
Through extensive experimentation and evaluation, we demonstrate the superior anomaly detection capabilities of these transformed classifiers. Their ability to accurately distinguish between normal and anomalous events provides valuable insight into potential new physical phenomena and ensures a robust framework for data analysis at the LHC.
Our results highlight the potential of reusing established classifiers for anomaly detection in particle physics. This approach not only extends the applicability of these models, but also improves the efficiency and accuracy of anomaly detection, facilitating the discovery of rare and interesting events in high-energy physics experiments.
| Collaboration / Activity | no collaboration, see comments |
|---|
