Erlangen 2026 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
T: Fachverband Teilchenphysik
T 58: Searches/BSM III
T 58.6: Talk
Wednesday, March 18, 2026, 17:30–17:45, KH 02.018
Exploring OmniJet-α for Model-Agnostic Anomaly Detection with CATHODE — •Salome Fresenbet1, Gregor Kasieczka1, Louis Moureaux1, Anna Hallin1, Joschka Birk1, Humberto Reyes Gonzalez2, and Soumya Shaw3 — 1University of Hamburg — 2RWTH Aachen University — 3Saarland University
Model-agnostic anomaly detection has become an important complement to traditional, signal-driven searches for new physics at the LHC. One such method is CATHODE, which combines conditional density estimation in invariant-mass sidebands with a weakly supervised classifier in the signal region. In this talk, we present the integration of the jet foundation model OmniJet-α into CATHODE by training it to distinguish signal events from background events modeled by the generative component of CATHODE. We evaluate this approach on a dijet dataset with beyond-the-Standard-Model (BSM) benchmark resonance signals, which we use as a realistic anomaly-detection scenario. The aim of this ongoing study is to quantify the impact of jet foundation models on CATHODE's performance and how they may enhance the sensitivity of future model-agnostic searches.
Keywords: BSM; Anomaly Detection; Machine Learning; Foundation Models