Astrophysical Neutrinos with IceCube: Deep Learning in Deep Ice

High-energy neutrinos have been recognized as unique tracers of cosmic ray accelerators, being produced in hadronic interactions and propagating away from the sources unattenuated and undeflected.

The challenge for neutrino telescopes is to identify the faint signal of these neutrinos amidst a very high background, and to reconstruct the neutrino direction with the precision needed to localize sources.

With the announcements by IceCube of the discovery of the astrophysical neutrino flux in 2013 and the association of a blazar-type galaxy with neutrino emission in 2018, the era of neutrino astronomy is now close at hand.This project will seek to make high-confidence identification of astrophysical neutrino sources.

It will undertake multi-messenger analyses of realtime and archival data. It will aim to maximize discovery potential through the development of more precise reconstruction of neutrino events. Recent applications of deep learning to IceCube data have proven enormously powerful for cascade-type neutrino events.

This project will focus on developing deep learning applications for track-type events, which intrinsically have better pointing than cascades but which still rely on conventional reconstruction techniques.

These gains can be converted to immediate benefits by re-analysis of the existing IceCube data set, with the potential to newly resolve discrete sources of the astrophysical neutrino flux.