Speaker
Description
The Pierre Auger Observatory is the largest currently running detector, that studies the extensive air showers of ultra-high energy cosmic rays. In this contribution, we provide an overview of three machine-learning techniques used to improve the understanding of data measured by the surface detector of the Observatory. All of these methods use the spatial and temporal information contained in the shower footprint that is measured by the surface detector stations.
The first of these methods demonstrates the application of deep learning and large simulation datasets to reconstruct the energy of the impinging cosmic ray. This approach has the potential to reduce a composition bias when compared to the techniques that rely on fitting the lateral signal distribution.
One of the primary objectives of the Observatory is to understand the evolution of mass composition with energy. We can achieve this using observables such as the depth of the shower maximum and the number of muons in the shower. In the second study, Long Short-Term Memory and
convolutional neural networks are employed to determine the depth of the shower maximum. Furthermore, the neural networks allow for the examination of its development of the mean and standard deviation as a function of energy.
The third work focuses on estimating the depth of the shower maximum and the number of muons in the air shower using signals from the upgraded stations of the surface detector. Transformer networks are used for this purpose. Using simulated data, the study demonstrates the promising potential for an accurate reconstruction of the primary mass by combining the measurements of shower maximum and muon number.
