Speaker
Description
Among nature's most elusive particles, tau neutrinos are notoriously hard to detect, with only a few dozen high-purity events ever observed. The Deep Underground Neutrino Experiment (DUNE) is set to change this dramatically by using an intense neutrino beam and leveraging millimeter-scale resolution to statistically isolate hundreds of tau-neutrino interactions.
In this talk, I present studies evaluating DUNE's tau-neutrino discrimination capability using NuGraph, a graph neural network designed to classify charged-current interactions. By operating directly on detector "hit" information (minimal energy depositions along particle trajectories) NuGraph offers a lightweight architecture that minimizes dependence on high-level reconstruction.
Trained on Monte Carlo samples for both nominal and tau-optimized beam configurations, the model demonstrates strong identification performance, with signal-to-background separation projected to exceed 8$\sigma$ of significance (assuming tau-optimized exposure). Due to its hit-based approach, NuGraph is also expected to be well-suited for atmospheric neutrinos, where incoming particles span the full sky. We present preliminary studies on atmospheric samples indicating that NuGraph maintains flavor-separation performance comparable to the beam configuration. Given that atmospheric neutrinos will be available early in DUNE's operation, this approach provides a promising pathway for immediate physics results in the experiment's initial phase.
| I read the instructions above | Yes |
|---|