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 97.1: Vortrag

Freitag, 4. April 2025, 09:00–09:15, VG 2.102

Hit-Filtering with Graph Neural Networks for Tracking at Belle II — •Greta Heine, Giacomo De Pietro, and Torben Ferber — Karlsruher Institut für Technologie (KIT), Karlsruhe, Deutschland

Over the next few years, the Belle II Experiment will increase its instantaneous luminosity, which will also lead to a significant increase in the beam background, affecting the efficiency of both online and offline tracking algorithms. To overcome this challenge and to facilitate the identification of displaced vertices for the discovery of new physics phenomena, Belle II needs more robust tracking algorithms.

Graph Neural Networks (GNNs) are a powerful class of machine learning models capable of adapting to irregular geometries and modeling complex relationships within detector hits. In this work, GNNs are used to filter background hits in the Belle II Central Drift Chamber based on edge classification using detector-level information. By filtering the background hits, both the track fitting performance as well as the computational efficiency can be improved at high background levels.

This talk will present the performance of this filtering approach for offline tracking algorithms on both simulated and real data, showing significant improvements in tracking efficiency and robustness under varying background conditions.

Keywords: tracking; graph neural networks; edge classification; Belle II; drift chamber