Parts | Days | Selection | Search | Updates | Downloads | Help
AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz
AKPIK 3: Machine Learning in Particle- and Astroparticle Physics
AKPIK 3.2: Talk
Thursday, April 3, 2025, 16:30–16:45, Theo 0.134
Searching for Ultra-High Energy Photons applying Machine Learning Methods Using the Surface Detector of the Pierre Auger Observatory — •Fiona Ellwanger for the Pierre-Auger collaboration — KIT, Karlsruhe, Germany
Identifying sources of cosmic rays is challenging, as the charged particles are deflected by magnetic fields and do not point back to their sources. Neutral particles, such as ultra-high energy (UHE) γ’s will point directly to their sources, unless they interact in the interstellar medium or are absorbed. Cosmic ray detectors such as the 3000 km2 surface array of the Pierre Auger Observatory are capable of observing UHE γ’s above 1018 eV. With increasing energy, their mean free path allows probing extragalactic sources up to a few Mpc. Unlike cosmic rays, photon-induced showers are almost purely electromagnetic. Different methods like BDTs and air-shower Universality have been previously applied to the search of γ’s at different energy ranges. Although no UHE γ’s have been found, the obtained bounds of the fluxes provide crucial constraints on cosmic-ray acceleration models.
Neural networks have the potential to improve discriminating variables, enhancing the sensitivity to even lower fluxes. In this work, we present a convolutional neural network designed to distinguish between simulated UHE photon and proton showers. We evaluate it on an independent test set, assessing its sensitivity and robustness to systematic uncertainties, including broken stations, detector aging, and noise. These steps aim to validate the network for application to the measured events.
Keywords: cosmic ray; ultra-high energy; photon; convolutional neural network