Date: Tuesday 10 Feb 2026 – 15:00 (Europe/London)
Speaker: Felix Yu, 4th year PhD student at Harvard University
Abstract
Neutrino telescopes detect rare particle interactions originating from some of the most extreme environments in the Universe. They achieve this by instrumenting a cubic-kilometer volume of transparent medium with light sensors. Owing to their size and the prevalence of background events, these detectors produce enormous amounts of high-dimensional, highly variable data. Such characteristics pose major challenges for predicting event properties such as direction and energy, particularly with machine learning (ML) methods. In this talk, I will present an efficient point cloud transformer model designed to address these challenges. I will also discuss a self-supervised training strategy that shifts the majority of learning to real data, thereby reducing reliance on simulations and mitigating associated systematic uncertainties.
Biography
Felix is a 4th-year PhD student at Harvard working at the intersection of neutrino astrophysics and deep learning. His research focuses on developing novel methods and adapting cutting-edge AI techniques to neutrino physics, with the goal of accelerating fundamental scientific discovery.
