Neural network based event selection on FPGA for COMET Phase-Ihttp://www-comet.kek.jp/COMET5/presentations/neural-network-based-event-selection-on-fpga-for-comet-phase-ihttp://www-comet.kek.jp/COMET5/@@site-logo/logo.png
Neural network based event selection on FPGA for COMET Phase-I
M. Miyataki1, Y. Fujii2, Y. Nakazawa3, K. Ueno1, M. Lee4, C. Yamada1, C. Wu1, S. Sun1
1 Osaka University, Department of Physics, Toyonaka, Japan 2 Imperial College London, Department of Physics, London, United Kingdom 3 High Energy Accelerator Research Organization (KEK), The Institute for Particle and Nuclear Studies, Tsukuba, Japan 4 Sungkyunkwan University, Department of Physics, Suwon, Republic of Korea
A neural network based online event selection is being developed for the COMET (COherent Muon To Electron Transition) Phase-I experiment. This experiment employs the world’s highest intensity muon beam and a detector system comprising a drift chamber and a trigger hodoscope. Such a highintensity muon beam would generate significant backgrounds, leading to high trigger rates. Thus, we are developing a machine learning-based online classification algorithm on FPGA, that utilizes information from both the drift chamber and the trigger hodoscope. The neural network is constructed using COMET Phase-I Monte Carlo simulation data and will be implemented on an FPGA board developed for COMET Phase-I. We are investigating the classification performance and latency of this algorithm. We plan to present its expected trigger efficiency and contribution to the experimental sensitivity.
