Ústav technické a experimentální fyziky Institute of Experimental and Applied Physics

Neutrino Characterisation using Convolutional Neural Networks in CHIPS water Cherenkov detectors

NázevTitle
Neutrino Characterisation using Convolutional Neural Networks in CHIPS water Cherenkov detectorsNeutrino Characterisation using Convolutional Neural Networks in CHIPS water Cherenkov detectors
Druh výsledkuResult type
Článek v časopiseJournal article
AutořiAuthors
J. Tingey, S. Bash, J. Cesar, T. Dodwell, S. Germani, P. Koojiman, P. Mánek, M. Ozkaynak, A. Perch, J. Thomas, L. Whitehead
DOIDOI
10.1088/1748-0221/18/06/P06032
Časopis / citaceJournal / citation
Journal of Instrumentation. 2023, 2023 (18)(06), P06032. ISSN 1748-0221.
RokYear
2023
JazykLanguage
eng
WoSWoS
001084428300001
ScopusScopus
2-s2.0-85164222597
RIVRIV
RIV/68407700:21670/23:00366619!RIV24-MSM-21670___
ProjektProject
Institucionální podpora na rozvoj výzkumné org.Institucionální podpora na rozvoj výzkumné org.; Inženýrské aplikace fyziky mikrosvětaEngineering applications of microworld physics

AbstraktAbstract

This work presents a novel approach to water Cherenkov neutrino detector event reconstruction and classification. Three forms of a Convolutional Neural Network have been trained to reject cosmic muon events, classify beam events, and estimate neutrino energies, using only a slightly modified version of the raw detector event as input. When evaluated on a realistic selection of simulated CHIPS-5kton prototype detector events, this new approach significantly increases performance over the standard likelihood-based reconstruction and simple neural network classification.

This work presents a novel approach to water Cherenkov neutrino detector event reconstruction and classification. Three forms of a Convolutional Neural Network have been trained to reject cosmic muon events, classify beam events, and estimate neutrino energies, using only a slightly modified version of the raw detector event as input. When evaluated on a realistic selection of simulated CHIPS-5kton prototype detector events, this new approach significantly increases performance over the standard likelihood-based reconstruction and simple neural network classification.