Erlangen 2026 – scientific programme

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

T: Fachverband Teilchenphysik

T 59: Neutrino Astronomy III

T 59.4: Talk

Wednesday, March 18, 2026, 17:00–17:15, KS H C

Advanced Northern Tracks Selection using a Graph Convolutional Neural Network for the IceCube Neutrino Observatory — •Philipp Soldin, Shuyang Deng, Lasse Düser, Sönke Schwirn, Christopher Wiebusch, and Marco Zimmermann — 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 via Cherenkov radiation emitted by secondary charged particles, recorded by more than 5,000 digital optical modules equipped with photomultiplier tubes (PMTs). A central challenge for IceCube analyses is the efficient separation of muons produced in neutrino interactions from the dominant background of muons from cosmic-ray air showers. To address this challenge, the Advanced Northern Tracks Selection (ANTS) employs a two-stage, machine-learning-based event selection. A transformer-based autoencoder first performs dimensionality reduction of the time-resolved sensor data, followed by a deep graph convolutional neural network (GCNN) that explicitly exploits the irregular, node-like geometry of the IceCube detector. Compared to established selection methods, ANTS achieves a significant improvement in classification performance. This presentation examines the ANTS network architecture, training strategy, background-rejection capability, and computational efficiency.