Since its installation, the ODSL compute server has enabled numerous contributions from ORIGINS to the scientific community. It serves as an informal but instrumental testbed complementary with clusters like C2PAP for new methods in statistics and artificial intelligence. To be able to compete within machine learning research in modern physics, we propose an additional server for the next generation of scientific challenges where nothing short of bleeding-edge hardware is sufficient.
An example is the research of the authors of this proposal: We intend to develop the next generation of neutrino event reconstruction algorithms. We have shown that our methods using state-of-the-art Graph Neural Networks can massively outpace algorithms based on traditional methods. Upgraded hardware will allow us to explore new techniques that will shape the future of physics. Such novel methods can help find evidence of neutrino emission where there now is
none and turn current evidence into discoveries.
Speakers:
Lukas Heinrich
(TUM, ODSL Coordinator),
Martin Ha Minh
(TUM, ODSL),
Rasmus Ørsøe
(TUM),
Philipp Eller
(TUM, ODSL Fellow),
Oliver Schulz
(MPP, ODSL Fellow)