Neutrino reconstruction with Graph Neural Networks in KM3NeT/ORCA6

• Main Author: Shen Liang
• Format: Conference Object
• Language: unknown
• Published: Zenodo 2022
• Subjects: Orca
• Online Access: https://doi.org/10.5281/zenodo.6785196

KM3NeT is a series of neutrino telescopes under construction in the Mediterranean Sea. The detectors will consist of 3D arrays of 3’’ PMTs distributed on 115 vertical detection units (DUs), each containing 18 digital optical modules (DOMs), with each DOM hosting 31 PMTs. The ORCA detector is designed with a dense arrangement of DOMs to study GeV-scale atmospheric neutrino oscillations. An early configuration of this detector dubbed KM3NeT/ORCA6, with 6 DUs, has been deployed and operated for almost 2 years from January 2020 to November 2021. Traditional reconstruction algorithms developed for the KM3NeT detectors rely on fitting the observed light patterns to single charged particle hypotheses. In contrast, a new approach based on Graph Neural Networks (GNNs) aims to use machine learning methods to extract more information from the full event topology of neutrino interactions. This contribution will detail the current implementation of GNNs and compare with traditional reconstruction algorithms in the context of the first neutrino oscillation analyses performed with the KM3NeT/ORCA6 detector.