Speaker
Description
The LHCb experiment at CERN operates a fully software-based first-level trigger that processes 30 million collision events per second, with a data throughput of 4 TB/s. Real-time tracking—reconstructing particle trajectories from raw detector hits—is essential for selecting the most interesting events, but must be performed under tight latency and throughput constraints.
A key bottleneck in this process is the suppression of “ghost” tracks: false combinations of detector hits caused by noise or detector ambiguities. To meet these challenges, a dedicated neural network architecture has been developed and optimized for fast inference on GPUs. Its speed is critical—enabling high-quality tracking to run within the strict timing budgets of the trigger system. The network achieves very good physics performance with ghost rates below 20%. The techniques employed can be used for real-time ML in similarly demanding environments.