Parts | Days | Selection | Search | Updates | Downloads | Help
T: Fachverband Teilchenphysik
T 70: Experimental Methods (general) 3
T 70.2: Talk
Wednesday, March 23, 2022, 16:30–16:45, T-H29
Convolutional Networks and Deep Learning at the Belle II Experiment — •Johannes Bilk, Sören Lange, Katharina Dort, Stephanie Käs, and Timo Schellhaas — Justus-Liebig-Universität, Gießen,Germany
The Belle II pixeldetector (PXD) has a trigger rate of up to 30 kHz for 8 M pixels. Its proximity to the interaction point allows it to detect exotic highly ionizing particles such as antideuterons, magnetic monopoles, stable tetraquarks or pions with small transverse momenta < 100 MeV. Those particles leave no tracks in the outer parts of the Belle II detector, and thus their pixel data may be deleted online as part of background suppression. In this contribution, we evaluate the performance of a machine learning algorithm to identify slow pions only on the basis of pattern recognition of pixel cluster structures. We employ convolutional neural networks with different kernel configurations and use images of 9x9 pixel matrices as input. On the long term such image recognition techniques could provide a rescue mechanism for the pixel data before they are erased. Results on accuracy and sensitivity are presented.