Using Convolutional Neural Networks to Reconstruct Energy of GeV Scale IceCube Neutrinos

TL;DR

This paper introduces a CNN-based method for reconstructing the energy of 10-GeV neutrino events in IceCube, achieving improved resolution and faster processing compared to traditional likelihood approaches.

Contribution

The study presents a novel CNN approach for neutrino energy reconstruction in IceCube, enhancing accuracy and efficiency over existing likelihood-based methods.

Findings

CNN achieves competitive energy resolution.

Reconstruction is faster than traditional methods.

Improved neutrino energy measurement for oscillation studies.

Abstract

The IceCube Neutrino Observatory, located under 1.4 km of Antarctic ice, instruments a cubic kilometer of ice with 5,160 optical modules that detect Cherenkov radiation originating from neutrino interactions. The more densely instrumented center, DeepCore, aims to detect atmospheric neutrinos at 10-GeV scales to improve important measurements of fundamental neutrino properties such as the oscillation parameters and to search for non-standard interactions. Sensitivity to oscillation parameters, dependent on the distance traveled over the neutrino energy (L/E), is limited in IceCube by the resolution of the arrival angle (which determines L) and energy (E). Event reconstruction improvements can therefore directly lead to advancements in oscillation results. This work uses a Convolutional Neural Network (CNN) to reconstruct the energy of 10-GeV scale neutrino events in IceCube, providing results with competitive resolutions and faster runtimes than previous likelihood-based methods.