Overview of Machine Learning Applications at the Pierre Auger Observatory

The complex spatio-temporal information from shower footprints, comprised of particle arrival times and traces measured by water-Cherenkov detectors, is challenging to analyse with traditional methods but well-suited for machine learning (ML) based analyses. In this contribution, we provide an overview of the ML applications developed to leverage the high event statistics acquired by the Pierre Auger Observatory. In the context of the energy spectrum, a neural network approach for energy reconstruction has demonstrated potential in reducing composition biases in the energy estimator. A notable application for mass composition is the indirect prediction of the depth of the maximum shower development, Xmax, which extends the energy range of previous analyses into unexplored higher energies. Aligned with AugerPrime, the ongoing upgrade of the Observatory, the impact of enhanced electronics and scintillation detectors was explored via simulations. Both transformers and convolutional networks perform better at the reconstruction of mass-composition sensitive observables like Xmax and the muon number, demonstrating the benefits of the Observatory’s upgrade.