Karlsruhe 2024 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

T: Fachverband Teilchenphysik

T 41: Trigger+DAQ 1

T 41.4: Talk

Tuesday, March 5, 2024, 16:45–17:00, Geb. 30.23: 3/1

Anomaly detection for the level 1 trigger system of the CMS experiment — •Sven Bollweg, Gregor Kasieczka, Karim El Morabit, Susan Sefidrawan, and Artur Lobanov — Universität Hamburg, Hamburg, Deutschland

There exist strong hints for the existence of physics beyond the standard model (BSM). At the CMS experiment only events passing the first selection step, the level 1 (L1) trigger, are recorded and available for further analysis. Assuming that BSM events differ from standard model (SM) events, a trigger selection targeting BSM events could be based on the detection of differing, i.e. anomalous, properties instead of criteria predicted by specific BSM models.

This talk discusses a neural network based implementation of such an anomaly detection trigger. An autoencoder (AE) network is trained to reproduce typical collision events. It is found that the reconstruction quality of anomalous events, such as BSM events or rare SM events, is decreased. The reproduction quality can be used as a basis to identify anomalous events, which could be BSM events. The integration of the AE into the existing L1 hardware and avoidance of overlap with the existing triggers presents additional challenges.

Keywords: Trigger; Machine Learning; Autoencoder; Level 1; Anomaly Detection