Robust Neural Particle Identification Models

Abstract The volume of data processed by the Large Hadron Collider experiments demands sophisticated selection rules typically based on machine learning algorithms. One shortcomings these approaches is their profound sensitivity to biases in training samples. In case particle identification (PID), this might lead degradation efficiency for some decays not present dataset due differences input kinematic distributions. talk, we propose a method Common Specific Decomposition that takes into account individual and possible misshapes disentangling common decay specific components feature set. We show proposed approach reduces rate PID algorithms reconstructed LHCb detector.