Erlangen 2026 – scientific programme

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

T: Fachverband Teilchenphysik

T 59: Neutrino Astronomy III

T 59.3: Talk

Wednesday, March 18, 2026, 16:45–17:00, KS H C

Approximating Photon Propagation in Ice Using Generative Neural Networks — •Amith Ashwath Narayan for the IceCube collaboration — Technical University of Munich

The Precision Optical Calibration Module (POCAM) is an isotropic, self-monitored calibration device. As part of the IceCube upgrade an extension of the IceCube detector located at the geographical South Pole, POCAMs are being installed to tackle existing optical detector systematics with higher precision. Estimating these detector systematic uncertainties requires parsing a multidimensional parameter space, which is computationally intensive, therefore it is infeasible. Since an analytical approximation with sufficient precision does not exist, we employ machine learning: by sparsely sampling the parameter space and using a neural network to interpolate between simulated points. In this project, the focus is specifically on scattering coefficient, training the network to generate the corresponding detection optical module (DOM)-response histograms and total photon counts. The talk will focus on the neural-network architecture and its performance in generating DOM-response histograms and photon counts for the scattering coefficient, with the method being in principle extendable to the remaining detector systematics.

Keywords: IceCube Upgrade; Optical calibration; Scattering coefficient; Photon propagation; Generative neural networks