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

T: Fachverband Teilchenphysik

T 94: Trigger+DAQ 3

T 94.9: Vortrag

Donnerstag, 7. März 2024, 18:00–18:15, Geb. 30.23: 3/1

Improving the Belle II Neural Track Trigger with Deep Neural Networks — •Timo Forsthofer^{1, 2}, Christian Kiesling^{1, 2}, Simon Hiesl^{1, 2}, and Kai Unger³ — ¹Max Plack Institute for Physics — ²Ludwig-Maximilians-Universität München — ³Karlsruhe institute of Technology

The Neural Track Trigger at Belle II is presently running, due to time limitations in the FPGA hardware, as a single hidden layer fully connected neural network. With rising background due to increased luminosity the performance during the last running period in 2022 has suffered both in resolution and background rejection.

Based on new FPGA hardware, which is now accessible, both the input and the structure of the neural networks can be redesigned. Most importantly, the neural network structure can be enlarged to a deep learning architecture.

We report here on improved performance by testing different network architectures and implementing new ways of preprocessing the inputs to the neural networks.

Keywords: Deep Neural Networks; Trigger; Belle II; FPGA