Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz

AKPIK 4: Deep Learning

AKPIK 4.8: Vortrag

Donnerstag, 24. März 2022, 18:00–18:15, AKPIK-H13

Event-by-event estimation of high-level observables with data taken by the Surface Detector of the Pierre Auger Observatory using deep neural networks — •Steffen Hahn¹, Markus Roth¹, Darko Veberic¹, David Schmidt¹, Ralph Engel¹, and Brian Wundheiler² — ¹KIT, IAP, Germany — ²UNSAM, ITEDA, Argentina

Probing physics beyond the scales of human-made accelerators with cosmic rays requires accurate estimation of high-level observables, such as the energy of the primary particle or the maximum of the shower depth. Measurements of the shower cascade, however, consist mainly of various, hard-to-interpret time signals which potentially contain non-trivial correlations. Deep neural networks are a convenient way to tackle such a problem in a general way.

The shower footprint measured by the surface detector of the Pierre Auger Observatory provides us with time slices of the ground signal of a shower cascade. This gives us an ideal test bed to determine the quality of network based reconstruction methods compared to that of regular analysis methods. However, a caveat of this approach is that the networks must be trained on Monte-Carlo simulations. Since present hadronic interaction models for energies beyond 10 EeV are extrapolations there are discrepancies between simulations and real data for which we have to correct for.

Here, we present a multi-purpose architecture and correction-method to predict high-level observables on measured data as well as physics results.