Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
HK: Fachverband Physik der Hadronen und Kerne
HK 27: Computing I
HK 27.1: Vortrag
Mittwoch, 12. März 2025, 14:00–14:15, SR Exp1A Chemie
Machine Learning Algorithms for Pattern Recognition with the PANDA Barrel DIRC — •Yannic Wolf1,2, Roman Dzhygadlo1, Klaus Peters1,2, Georg Schepers1, Carsten Schwarz1, and Jochen Schwiening1 for the PANDA collaboration — 1GSI Helmholtzzentrum für Schwerionenforschung GmbH, Darmstadt — 2Goethe-Universität Frankfurt
Precise and fast hadronic particle identification (PID) is crucial to reach the physics goal of the PANDA detector at FAIR. The Barrel DIRC (Detection of Internally Reflected Cherenkov light) is a key detector for the identification of charged hadrons in PANDA. In comparison to other Ring Imaging Cherenkov detectors, the hit patterns observed with DIRC counters do not appear as rings on the photosensor plane but as complex, disjoint 3D-patterns.
Using the recent advances in machine learning (ML) algorithms, especially in the areas of image recognition and generative AI, we have studied ML PID algorithms for the PANDA Barrel DIRC. Several network implementations were found to be capable of reaching a performance comparable to conventional methods, but only if the network is trained for each particle angle and momentum. To make a trained network usable for different points in phase space, and to optimize the training process and PID performance, we varied the data input structures, increased the parameter space, and included normalizing flow-based generative models in the study. We will show a comparison of the performance of different ML methods to conventional algorithms and discuss the impact on the PANDA Barrel DIRC.
Keywords: Machine Learning; AI; PANDA; DIRC