Machine Learning-based Energy Reconstruction for Water-Cherenkov detectors

Abstract

Hyper-Kamiokande (Hyper-K) is a proposed next generation underground water Cherenkov (WCh) experiment. The far detector will measure the oscillated neutrino flux from the long-baseline neutrino experiment using 0.6 GeV neutrinos produced by a 1.3 MW proton beam at J-PARC. It has a broad program of physics and astrophysics mainly focusing on the precise measurement of the lepton neutrino mixing matrix and the CP asymmetry. The unoscillated neutrino flux will be measured by an intermediate WCh detector. One of the proposed designs is the Tokai Intermediate Tank for the Unoscillated Spectrum (TITUS). WCh detectors are instrumented with photomultipliers to detect the Cherenkov light emitted from charged particles which are produced by neutrino interactions. The detection of light is used to measure the energy, position and direction of the charged particles. We propose machine learning-based methods to reconstruct the energy of charged particles in WCh detectors and present our results for the TITUS configuration.