ODSL Forum: GraphNeT & Its Applications in The IceCube Neutrino Observatory

by Rasmus Ørsøe (TUM)

Friday 21 Apr 2023, 15:00 → 16:00 Europe/Berlin

Basement Seminar Room (Origins Building)

Abstract: This talk will explain why the geometric time series data from neutrino telescopes are challenging in a deep learning context, motivate why the abstract machine learning paradigm graph neural networks(GNNs) are a solution to some of those issues and introduceGraphNeT, an open-source python framework for development and applications of GNNs in neutrino telescopes. The talk will show how GNNs from GraphNeT applied to the IceCube Neutrino Observatory can serve as drop-in replacements to existing physics analyses (and likely improve them) and how they have enabled new studies that was previously out of reach.  

ODSL Seminar GraphNeT.pdf