Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
T: Fachverband Teilchenphysik
T 7: Data, AI, Computing, Electronics I
T 7.1: Vortrag
Montag, 16. März 2026, 16:15–16:30, KH 00.024
Understanding and Expanding the Transformer-Based Neural Network to Analyze Extensive Air Showers at the Pierre Auger Observatory — •Ronja Westphalen, Maximilian Straub, Alex Reuzki, Niklas Langner, Josina Schulte, and Martin Erdmann — RWTH Aachen University, Aachen, Germany
Determining the mass of ultra-high-energy cosmic rays is crucial to understanding their origin. Ground-based detectors, such as the Pierre Auger Observatory, measure extensive air showers and collect multidimensional, time- and location-dependent signals that contain information on the primary particle. Analyzing these complex signals with a Transformer-based neural network to reconstruct mass-dependent observables, such as the depth of the shower maximum Xmax and the muon content Rµ, from surface detector (SD) measurements at the Pierre Auger Observatory has been shown to be successful.
This study investigates the working principle of the Transformer in detail. We examine the attention mechanism in the time trace and in spatial analysis to understand how effectively features are reconstructed from these complex measurements. Additionally, we assess the agreement between the data and simulations in the latent network space to determine whether known discrepancies of hadronic interaction models are inherently apparent in the network response. Finally, we want to extend the network by the radio detector, enabling the evaluation of SD measurements in combination with the radio signal for the AugerPrime setup.
Keywords: Transformer; ultra-high-energy cosmic rays; extensive air showers; attention; hadronic interaction model