Particle identification in the Belle II experiment involves utilizing information from various sub-detectors to classify six different species of charged particles: electrons, muons, charged pions, charged kaons, protons, and deuterons. Previous studies have demonstrated that directly adding log-likelihoods from each detector for each hypothesis is not an optimal use of available information since poorly calibrated detectors can hurt overall particle identification performance. To address these issues, we study different approaches that involve assigning to the individual contributions different weights, depending on the region of the phase space under study. Machine learning tools are employed in order to optimize the weights and study the possible improvements in the performance.
Identificazione delle particelle nell'esperimento Belle II implica l'utilizzo delle informazioni provenienti da vari sotto-rivelatori per classificare sei diverse specie di particelle cariche: elettroni, muoni, pioni carichi, kaoni carichi, protoni e deuteroni. Studi precedenti hanno dimostrato che l'aggiunta diretta dei log-likelihood di ciascun rivelatore per ciascuna ipotesi non rappresenta un utilizzo ottimale delle informazioni disponibili, poiché i rivelatori scarsamente calibrati possono compromettere le prestazioni complessive dell'identificazione delle particelle. Per affrontare tali problematiche, studiamo diverse approcci che prevedono l'assegnazione di diversi pesi ai contributi individuali, a seconda della regione dello spazio delle fasi in studio. Strumenti di apprendimento automatico vengono impiegati al fine di ottimizzare i pesi e studiare i possibili miglioramenti nelle prestazioni.
Optimization of the PID algorithms at the Belle II Experiment
BAVARCHEE, ALI
2022/2023
Abstract
Particle identification in the Belle II experiment involves utilizing information from various sub-detectors to classify six different species of charged particles: electrons, muons, charged pions, charged kaons, protons, and deuterons. Previous studies have demonstrated that directly adding log-likelihoods from each detector for each hypothesis is not an optimal use of available information since poorly calibrated detectors can hurt overall particle identification performance. To address these issues, we study different approaches that involve assigning to the individual contributions different weights, depending on the region of the phase space under study. Machine learning tools are employed in order to optimize the weights and study the possible improvements in the performance.File | Dimensione | Formato | |
---|---|---|---|
Bavarchee_Ali.pdf
accesso aperto
Dimensione
6.99 MB
Formato
Adobe PDF
|
6.99 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/47361