Speaker
Description
This work focuses on adapting LHCb’s existing machine learning approach to electron identification to the upgraded conditions of Run 3. The LHCb experiment at CERN focuses on precision measurements in heavy-flavour physics, where efficient particle identification (PID) is crucial. During the last LHC Long Shutdown (2019-
2022) a major upgrade to the detector was done, including a fully-software trigger, new tracking detectors and an update of the RICH (Ring-Imaging Cherenkov) subdetectors, in order to be able to work at increased data rates. As a consequence of all these upgrades, the re-optimization of PID algorithms is then required. One of these PID algorithms is the ProbNNe algorithm, a neural network-based binary classifier, which estimates how likely a reconstructed particle is actually an electron. The model takes as input both PID and tracking information from the detector and outputs the corresponding probability. The information that serves as training samples are Monte Carlo (MC) simulations that emulate the conditions of the 2024 data-taking. After the training and tuning processes, the model is tested on independent MC samples as well as on real, recorded data from 2024.
