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

T: Fachverband Teilchenphysik

T 77: Deep Learning III

T 77.2: Vortrag

Donnerstag, 28. März 2019, 16:15–16:30, H06

Deep Learned Calorimetry with the CALICE AHCAL Technological Prototype — •Erik Buhmann and Gregor Kasieczka for the CALICE-D collaboration — Universität Hamburg, Institut für Experimentalphysik, Luruper Chaussee 149, 22761 Hamburg

The Analog Hadron Calorimeter (AHCAL) Technological Prototype 2018, developed by the CALICE collaboration for a future linear collider experiment, is a highly granular calorimeter consisting of roughly 22,000 individual scintillator tiles. The prototype underwent test beam in May, June and October 2018 at the SPS. Hit energies and hit times in each individual channel are recorded and can be processed into 3D images of single events. In this study we use Deep Learning algorithms to analyze those images.

Convolutional Neural Nets (CNNs) are used for two separate tasks, for energy reconstruction as well as for particle classification. The training of the neural networks are performed with test beam data as well as using a Monte Carlo simulation. Studies of the energy reconstruction performance with different CNN architectures, preprocessing and data augmentation are presented. A comparison between a cut based approach and the CNN performance for particle classification is discussed.