How Neutrino Astronomy May Inform the Next-Generation of IACT Event Reconstruction Techniques

by Jarred Green (Max Planck Institute for Physics)

Tuesday Oct 28, 2025, 12:00 PM → 1:00 PM Europe/Madrid

C7b/058 - Seminar (IFAE Main Building C7b)

Conventional gamma-ray event reconstruction for Imaging Atmospheric Cherenkov Telescopes (IACTs) relies on a multi-stage pipeline that reduces air shower images to a set of hand-engineered parameters. This process of image cleaning and parameterization may discard potentially valuable information, especially the detailed temporal structure of the Cherenkov signal. In this study, we explore an end-to-end approach, inspired by techniques from neutrino astronomy, that learns directly from low-level detector data. We apply deep learning models for the first time directly to the calibrated 30 ns waveform signals from the MAGIC telescopes, bypassing the standard image processing chain entirely. To naturally accommodate the irregular camera geometry and asynchronous pixel readouts, we leverage Graph Neural Networks (GNNs) and transformers via the GraphNeT framework, a pipeline originally developed for Neutrino observatories like IceCube. Our preliminary findings, presented in this talk, show that these generalizable techniques can effectively classify events and reconstruct gamma-ray arrival direction from raw data. This highlights a potentially powerful alternative to traditional methods, demonstrating the benefits of when techniques can hop between domains.