Erlangen 2026 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
T: Fachverband Teilchenphysik
T 5: Methods in Particle Physics I
T 5.1: Talk
Monday, March 16, 2026, 16:15–16:30, KH 00.020
Machine Learning Methods for Charged-Particle Identification at Belle II — •Robert Mundzeck1,2, Martin Bartl1, Hans-Günther Moser1, and Stefan Wallner1 — 1Max Planck Institute for Physics, Garching, Germany — 2Technical University of Munich
We present recent advances in charged-particle identification at the Belle II experiment at KEK, Japan. So far, particle identification at Belle II has relied on a likelihood-based method that combines information from six subdetectors to distinguish between different particle species. Recently, a neural-network-based classifier has been developed that complements and improves upon this conventional approach. We report on a study comparing the neural network performance to an alternative classifier based on boosted decision trees. We also present the approaches of further optimization of the neural network, aimed at optimizing particle identification performance at Belle II.
Keywords: Belle II; Charged-Particle Identification; Machine Learning; Neural Network; Boosted Decision Tree