Parts | Days | Selection | Search | Updates | Downloads | Help
AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz
AKPIK 3: Machine Learning in Particle- and Astroparticle Physics
AKPIK 3.1: Talk
Thursday, April 3, 2025, 16:15–16:30, Theo 0.134
A Hybrid Approach for Optimizing Background Simulations in IceCube — •Simon Koch, Christian Haack, and Benedikt Mayer — Erlangen Centre for Astroparticle Physics - ECAP, FAU Erlangen-Nürnberg
The IceCube Neutrino Observatory detects high-energy cosmic neutrinos by observing Cherenkov radiation emitted from secondary particles, such as muons, produced in neutrino interactions. A key challenge in detecting cosmic neutrinos is the large background of cosmic-ray induced muons, which has to be reduced by a factor of ∼ 107.
Thus a large sample of background events has to be simulated in order to accurately estimate the background reduction efficiency. The computationally most expensive part of the simulation chain is the propagation of Cherenkov photons, induced by the muon energy losses.
In this work we develop a hybrid simulation approach that combines traditional simulation methods with a surrogate model. Our surrogate model predicts the probability of cosmic ray induced muons surviving the background reduction process based on the muon energy loss information. This approach ensures that computational resources required for the photon propagation of the background events are better spent on statistically rare events, which have a high chance of surviving the background reduction. For a given sample size of background events, we are thus able to reduce the statistical uncertainty of the estimated background reduction efficiency.
Keywords: machine learning; IceCube; simulation; neutrino; background reduction efficiency