Machine learning for particle identification in the LHCb detector

LHCb experiment is a specialised b-physics experiment at the Large Hadron Collider at CERN. It has a broad physics program with the primary objective being the search for CP violations that would explain the matter-antimatter asymmetry of the Universe. LHCb studies very rare phenomena, making it necessary to process millions of collision events per second to gather enough data in a reasonable time frame. Thus software and data analysis tools are essential for the success of the experiment. Particle identification (PID) is a crucial ingredient of most of the LHCb results. The quality of the particle identification depends a lot on the data processing algorithms. This dissertation aims to leverage the recent advances in machine learning field to improve the PID at LHCb. The thesis contribution consists of four essential parts related to LHCb internal projects. Muon identification aims to quickly separate muons from the other charged particles using only information from the Muon subsystem. The second contribution is a method that takes into account a priori information on label noise and improves the accuracy of a machine learning model for classification of this data. Such data are common in high-energy physics and, in particular, is used to develop the data-driven muon identification methods. Global PID combines information from different subdetectors into a single set of PID variables. Cherenkov detector fast simulation aims to improve the speed of the PID variables simulation in Monte-Carlo.