SMuK 2023 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
T: Fachverband Teilchenphysik
T 34: ML Methods II
T 34.5: Vortrag
Dienstag, 21. März 2023, 18:00–18:15, HSZ/0405
Resonant anomaly detection without background sculpting — •Manuel Sommerhalder1, Gregor Kasieczka1,2, Tobias Quadfasel1, Anna Hallin3, and David Shih3 — 1Institut für Experimentalphysik, Universität Hamburg, 22761 Hamburg, Germany — 2Center for Data and Computing in Natural Sciences (CDCS), 22607 Hamburg, Germany — 3NHETC, Dept. of Physics and Astronomy, Rutgers University, Piscataway, NJ 08854, USA
Anomaly searches are a class of machine learning--based methods to search for new phenomena without relying on specific signal and background models. They provide a promising complement to the typically model-dependent searches for physics beyond the standard model at the LHC. Resonant anomaly detection methods, such as CATHODE, make use of the assumptions of a signal being localized in one feature and have demonstrated great performance in terms of classifying new physics signals on simulation-based studies. However, they are prone to background sculpting in the case of input features being correlated with the resonant one and thus can ultimately impair a background estimation via the bump hunt. We thus propose Latent CATHODE (LaCATHODE), a new technique for resonant anomaly detection, which moves the features into a decorrelated latent space. Using the LHC Olympics R&D dataset, we observe that LaCATHODE leaves the background unsculpted while retaining much of the signal extraction performance of the original CATHODE approach.