Göttingen 2025 – wissenschaftliches Programm

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

T: Fachverband Teilchenphysik

T 97: Data, AI, Computing, Electronics IX (AI-based Object Reconstruction)

T 97.2: Vortrag

Freitag, 4. April 2025, 09:15–09:30, VG 2.102

End-to-End Multi-Track Reconstruction using Graph Neural Networks at Belle II — •Lea Reuter, Giacomo De Pietro, and Torben Ferber — Institute of Experimental Particle Physics, Karlsruhe Institute of Technology, Karlsruhe, Germany

Displaced vertices are an important signature in Standard Model analyses involving K_S and many searches for New Physics. However, the current Belle II tracking algorithm falls short when dealing with particles that decay after a large distance, resulting in a decrease in tracking efficiency with increasing displacement.

In this work, we show a novel track finding algorithm that combines the Object Condensation algorithm with Graph Neural Networks. This approach simultaneously identifies all tracks in an event and determines their respective parameters. Additionally, we integrated the new track finding algorithm into the Belle II analysis software framework.

Our results show significant reconstruction improvements of more than 50% for a long-lived particle within the GeV mass range and a lifetime of 10 cm in comparison to the existing Belle II track finding algorithm. This improvement is achieved while maintaining a similar efficiency and fake rate for prompt tracks originating from the interaction point.

Keywords: Tracking; Graph Neural Networks; Machine Learning; Drift Chamber; Reconstruction