AI and the World's Hard Problems · Physics · arXiv · 2026-10-20
Low-Latency Beam-Background Suppression for the BESIII Trigger System Using Machine Learning Trained on Experimental Data
Beam-induced backgrounds increase the trigger and data-acquisition rate at high-luminosity electron-positron colliders. The authors trained compact neural-network classifiers on BESIII experimental data, using 48 binary Level-1 trigger conditions, in one configuration based on tracking and time-of-flight and in another based on electromagnetic-calorimeter information. At matched hadron-event efficiency, the beam-background false-positive rate fell by 20–60% relative to the established trigger selection, and the network was shown to be suitable for low-latency FPGA implementation.