Speaker
Description
In 2023, KM3NeT observed a high-energy neutrino candidate with an estimated energy of greater than 100 PeV. The IceCube Neutrino Observatory, despite its substantially larger exposure, has not yet observed a neutrino of comparable energy. This motivates further efforts to enhance IceCube’s sensitivity at the highest energies. To improve the selection efficiency for high-energy neutrinos, we use machine learning techniques to enhance both event reconstruction and muon bundle rejection. Using graph neural networks (GNNs), we reconstruct the lateral spread of muons in an event and distinguish neutrinos from high-energy atmospheric muon bundles. Additionally, a Transformer-based neural network is used to reconstruct events, including direction, energy, and their event topology. Unlike previous likelihood-based reconstructions that assume a specific morphology, no prior assumptions are made before applying this reconstruction. We detail an enhanced event selection utilizing the improved background rejection and reconstruction methods.