Erlangen 2026 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

T: Fachverband Teilchenphysik

T 71: Data, AI, Computing, Electronics VII

T 71.7: Vortrag

Donnerstag, 19. März 2026, 17:45–18:00, KH 00.024

Improving Machine-Learning-Driven Anomaly Detection for New Physics Searches at Belle II — •Gianni Di Paoli, David Giesegh, and Thomas Kuhr — Ludwig-Maximilians-Universität München (LMU), München, Germany

Anomaly detection based on machine-learning techniques, using semi- or unsupervised methods, offers a complementary strategy to traditional theory-driven searches for New Physics beyond the Standard Model. Previous studies have provided a proof of principle by enhancing the visibility of simulated New Physics signals.

In this work, the performance of the existing network architectures is improved. For the autoencoder-based method, refinements of the network design lead to more stable reconstruction-error distributions, and thus to increased visibility of anomalies. For the density-estimation approach, alternative likelihood-estimation techniques and their behavior in high-dimensional feature spaces relevant for Belle II are investigated and implemented to improve sensitivity to anomalies.

Ongoing work focuses on further developing these techniques and preparing them for application to real Belle II collision data, including studies of their behavior under realistic detector and background conditions. This talk will present the current status of the algorithmic developments, and outline the next steps toward an operational anomaly-detection pipeline for model-independent New Physics searches at Belle II.

Keywords: Anomaly Detection; New Physics; Autoencoder; Density Estimation; Belle II