Improving event classification at KM3NeT with OrcaNet

Abstract

Neutrino telescopes such as KM3NeT are being built to detect these tiny, elusive particles. In the case of KM3NeT ORCA the aim is to determine the currently unknown neutrino mass hierarchy, which has far reaching implications for scientific research. Distinguishing between the different types of neutrino flavour interactions seen in the detector is critical for this goal. In order to achieve this, Deep-Learning algorithms such as the OrcaNet framework for KM3NeT are being developed and tested. This work consists of an exploration of the performance of this tool for the concrete case of event identification in KM3NeT, and its implications for determining the neutrino mass hierarchy. Here, clear evidence is presented that there is potential for event classification and identification beyond the current binary track-shower scheme, including up to 40% separation for electron neutrino charged current events.