AI and the World's Hard Problems · Physics · arXiv · 2026-10-19
Deep-learning-based low-energy trigger algorithms for the Hyper-Kamiokande experiment
The authors evaluated deep-learning trigger algorithms for Hyper-Kamiokande, a large water Cherenkov detector, aimed at neutrino events below 7 MeV. A supervised classifier and two anomaly-detection methods trained only on detector noise (an autoencoder and Manifold Projection-Diffusion Recovery) were compared with a hit-count trigger. For 3 MeV single electrons, the supervised model reached 76.7% signal efficiency versus 26.4% for the hit-count trigger, and Manifold Projection-Diffusion Recovery reached 35.4% at the same operating point, with GPU inference well below a millisecond per window.