Machine Learning in Particle Physics

Machine learning has played an integral role in supporting particle physics research, enabling scientists to analyze vast complex datasets generated by high-energy collisions. This approach helps identify and reconstruct elusive particles, such as the Higgs boson and pave the path to landmark discoveries.

In his talk, Sergei Gleyzer, co-director of the Alabama Center for the Advancement of Artificial Intelligence, from The University of Alabama, shares insights into the transformative role of machine learning in particle physics, accelerating discoveries at facilities like CERN’s Large Hadron Collider. Machine learning techniques have been used across a range of workflows within high-particle physics such as particle tracking, event classification, and event simulations. Sergie further describes a joint project with MathWorks titled “Graph Neural Networks for Realtime Top Quark Tagging”, that led to the creation of custom Graphical Neural Networks (GNN) deployed on a Compact Muon Solenoid (CMS) Trigger System, to enable the identification of subtle particle signatures that traditional algorithms might miss. This GNN, deployed on a real-time detection edge system, is intended to assist researchers to identify potential new discoveries within the vast amounts of data generated by detectors, 99.9% of which is discarded. The approach combines MATLAB and Deep Learning Toolbox for model quantization and deployment, maintaining high accuracy while meeting stringent real-time requirements.