000004204 001__ 4204
000004204 005__ 20240403134838.0
000004204 037__ $$aBELLE2-MTHESIS-2024-003
000004204 041__ $$aeng
000004204 100__ $$aNaveen Kumar Baghel
000004204 245__ $$aRecovery of merged π0,s from ECL images of the Belle II detector using CNN
000004204 260__ $$aMohali, India$$bIndian Institute of Science Education and Research, Mohali$$c2023
000004204 300__ $$a70
000004204 500__ $$aPresented on 01 05 2023
000004204 502__ $$aMSc$$bMohali, India, Indian Institute of Science Education and Research, Mohali$$c2023
000004204 520__ $$aThis study aims to utilize a Convolutional Neural Network (CNN) to retrieve merged π0 mesons lost in the Belle II experiment, where they appear as individual photons. The issue arises when dealing with high momentum π0 mesons, i.e., beyond 2 GeV in the Belle II experiment, as the shower produced by both the π0 meson and gamma appear indistinguish- able at the Electromagnetic Calorimeter (ECL) detector. Currently, reconstruction software is utilized to match photon pairs created by the π0 → γγ decay; however, the efficiency of this process can be affected by the γ produced by the rest of the events (ROEs), which mimic the signal. One of the most challenging tasks in particle physics research is accu- rately identifying and reconstructing subatomic particles. By the nature of the problem and its importance, accurate reconstruction of π0 mesons is crucial for identifying various B/D meson decays, including rare decays like D0 → γγ8.5×10−7,D0 → ρ0γ10−5, and D0 → φγ 10−5. These rare decays have dominant background arising from decays like D0 →Ksπ01.24×10−2,D0 →π0π08.26×10−4,andD0 →φπ01.17×10−3. The Convolutional Neural Networks performed reasonably well on a test dataset, which is identical to real scenarios, achieving an area under the curve (AUC) of 0.86 for the Precision-Recall curve. These results demonstrate the potential of machine learning (ML) algorithms and highlight areas for improvement in the current work to enhance the effi- ciency of identifying π0 particles with energy deposits in the ECL. The findings suggest that the ‘raw’ ECL images contain much more information than currently used expert- engineered features.
000004204 700__ $$aDr. Vishal Bhardwaj$$edir.
000004204 8560_ $$fnaveen.baghel@louisville.edu
000004204 8564_ $$uhttps://docs.belle2.org/record/4204/files/BELLE2-MTHESIS-2024-003.pdf
000004204 980__ $$aTHESIS