Speaker
Description
The Pierre Auger Observatory is the world's largest hybrid detection facility dedicated to the study of ultra-high-energy cosmic rays (UHECRs). At these extreme energies, cosmic rays offer a unique window into the most energetic astrophysical processes. However, since the UHECR flux is very low, they can only be indirectly observed through the extensive air showers they initiate when interacting with the Earth's atmosphere. By combining multiple detection techniques at the Observatory, the different parts of these cascades can be observed simultaneously, enabling complementary and cross-calibrated observations. A crucial piece of information for understanding the physics of UHECRs is their mass composition. The composition must be inferred from mass-sensitive shower observables, such as the depth of the shower maximum and the number of muons produced in the cascade, which stochastically encode the primary mass. While these observables can be measured directly by dedicated detector systems, such as the Fluorescence Detector and Underground Muon Detector of the Observatory, their operation is limited to specific conditions or regions and thus to reduced exposure. In contrast, the Surface Detector (SD) of the Observatory operates with nearly 100% uptime, recording the secondary particles at ground level. However, extracting mass-sensitive information from these complex spatio-temporal SD signals poses substantial analytical challenges. In this contribution, we present and summarize the current applications of machine learning (ML) techniques developed by the Pierre Auger Collaboration to fully exploit the SD data, focusing on neural-network-based reconstructions of high-level observables. By leveraging the large statistics of the SD, these ML-driven approaches extend composition studies to higher energies and improve the precision and robustness of mass reconstruction beyond the limitations of traditional analysis methods.