Göttingen 2025 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

T: Fachverband Teilchenphysik

T 97: Data, AI, Computing, Electronics IX (AI-based Object Reconstruction)

T 97.5: Talk

Friday, April 4, 2025, 10:00–10:15, VG 2.102

Advanced Northern Tracks Selection using a Graph Convolutional Neural Network for the IceCube Neutrino Observatory: Network Architecture — •Philipp Soldin, Philipp Behrens, Jakob Böttcher, Shuyang Deng, Lasse Düser, Philipp Fürst, Leon Hamacher, Michael Handt, and Christopher Wiebusch for the IceCube collaboration — RWTH Aachen University

The IceCube Neutrino Observatory is a large neutrino detector located in the ice at the geographic South Pole. It detects atmospheric and astrophysical neutrinos through the Cherenkov radiation emitted by secondary particles using over 5,000 photomultipliers (PMTs). One of the primary challenges is effectively distinguishing between muons induced by either neutrinos or air-showers. To address this, the Advanced Northern Tracks Selection (ANTS) employs a deep graph convolutional neural network (GCNN). This neural network takes advantage of the node-like structure of the PMT array's geometric arrangement and processes raw sensor data. By leveraging both local and global event features, the ANTS GCNN enhances classification performance. This presentation focuses on the architecture of the ANTS GCNN and evaluates its performance in rejecting background interference from air-shower-induced muons. We assess the accuracy and resolution of the reconstruction, and computational efficiency, showing significant improvements over traditional methods across various muon track topologies.

Keywords: IceCube; Advanced Northern Tracks Selection; Neutrino Astronomy; Deep Learning