Xavier Simó ; Stefan Wallner ; Stephan Paul

2023
Technische Universität München Munich

Abstract: The Belle II experiment aims to study the standard model (SM) of particle physics and to search for new physics (NP) beyond the standard model with unprecedented precision. This requires to accurately identify particles produced in these decays, necessitating the development of robust methods for particle identification (PID). This thesis presents a novel approach to charged particle identification using a neural network aiming to separate hadrons and leptons. We achieved a significant improvement in K/π separation performance. Subsequently, we developed an extended neural network for the simultaneous separation of electrons, muons, pions, kaons, protons, and deuterons. With this extended neural network, we achieved the same performance for K/π separation as with the specialized neural network. Furthermore, our approach outperforms the standard method for PID employed at Belle II, as well as another machine-learning based method developed to perform only lepton identification called lepton BDT. This is shown for binary classification, i.e separation of a pair of species. Furthermore, we show that multi-class classification, i.e. the separation of one species from a set of other species, comes with additional challenges. Also, for multi-class classification our neural network approach outperforms the other PID methods used at Belle II. In summary, in this work we have developed a universal neural network. This means that it is able to perform hadron and lepton identification with a better performance than all existing methods for particle identification at Belle II.

Note: Presented on 30 11 2023
Note: MSc