graphNeT
Graph neural networks for neutrino telescope event reconstruction
GraphNeT is a larger project to provide classification and reconstruction models for neutrino telescopes, based on graph neural networks. This type of NN architecture is very well suited to represent the data on an event-by-event basis without the need to resort to approximations or compression.
With my PhD students and collaborators from the NBI, we developed some very successful prototypes, that already today can outperform the classic algorithms in both accuracy and speed. A first publication is currently under review.
Further Information
Collaborators: Rasmus Orsoe (TUM), Martin Ha Minh, Dr. Andreas Sogard (NBI), Prof. Dr. Troels Petersen (NBI)
GitHub project page: https://github.com/icecube/graphnet
Citable resource: https://zenodo.org/record/6720189
Paper in preparation.