Heidelberg 2022 – wissenschaftliches Programm

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

T: Fachverband Teilchenphysik

T 41: Calorimeters 1

T 41.8: Vortrag

Dienstag, 22. März 2022, 18:00–18:15, T-H26

Convolutional Neural Networks for the Energy Reconstruction of ATLAS Liquid-Argon Calorimeter Signals — Anne-Sophie Berthold, Nick Fritzsche, •Christian Gutsche, Max Märker, Johann Christoph Voigt, and Arno Straessner — Institut für Kern- und Teilchenphysik, TU Dresden, Germany

In 2027, it is planned to start the High-Luminosity LHC, which will push the possibilities of research in particle physics with ATLAS to a new level. But since a higher trigger rate and more simultaneous collisions imply more pile-up the readout electronics of the detector will face a new challenge. The signal processing of the LAr Calorimeter is currently using an optimal filter algorithm which will reach its limits in performance with increasing overlapping signals. New approaches for energy-reconstruction are needed, and neural networks are promising candidates for such a task. While it is not hard to build a neural network which reconstructs energies reliably with a lot of trainable parameters, the problem is the limited availability of resources on the FPGAs which are foreseen for the digital signal processing.

In this talk, a possible solution for this task using convolutional neural networks (CNNs) will be presented. It will be shown how CNNs can be structured and trained in such a way that they will fit to the above-mentioned requirements. Special attention will be paid to the energy resolution for signals with a small temporal distance, having the pile-up at the HL-LHC in mind.