Erlangen 2026 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
T: Fachverband Teilchenphysik
T 7: Data, AI, Computing, Electronics I
T 7.6: Talk
Monday, March 16, 2026, 17:30–17:45, KH 00.024
Application of FiLM Neural Networks for π/K Separation in the PANDA Barrel DIRC — •Daniel Markhoff1,2, Roman Dzhygadlo2, Jochen Schwiening2, and Yannic Wolf2,3 — 1University of Edinburgh, Edinburgh, United Kingdom — 2GSI Helmholtzzentrum für Schwerionenforschung GmbH, Darmstadt, Germany — 3Goethe-Universität Frankfurt, Frankfurt, Germany
Machine learning techniques were investigated as an alternative approach to PID in the PANDA Barrel DIRC at FAIR. FiLM neural-network models were trained on simulated Geant4 photon-hit patterns and achieved >2σ π/K separation at 3.5 GeV/c. The models provide a significant reduction in computation time compared to the conventional time-based imaging reconstruction, while retaining competitive classification performance. These results indicate that ML-based PID has strong potential to complement or accelerate traditional DIRC reconstruction methods.
Keywords: PANDA; DIRC; particle identification; machine learning; FiLM networks