Optimization and deployment of an IA pipeline based on Graph Neural Networks for tack reconstruction in LHCb
Abstract:
Increases in instantaneous luminosity and detector granularity will increase the amount of data that has to be analyzed by high-energy physics experiments, whether in real time or offline, by an order of magnitude. In this context, Graph Neural Networks have received a great deal of attention in the community for the reconstruction of charged particles, because their computational complexity scales linearly with the number of hits in the detector.
We present a GNN reconstruction of LHCb’s vertex detector and benchmark its computational performance on both GPU and CPU architectures. A unique aspect of our work is the integration from the Python pipeline into LHCb's fully C++/Cuda GPU-based first-level trigger system, Allen, which performs at the rate of 20~MHz in the ongoing Run~3.
We present the first attempt to operate a GNN charged particle reconstruction in such a high-throughput environment using GPUs, and we discuss the pros and cons of the GNN and classical algorithms in a detailed like-for-like comparison.
Nabil Garroum is inviting you to a scheduled Zoom meeting.
Topic: ML Grooup Meeting at LPNHE
Time: Mar 28, 2024 02:00 PM Paris
Join Zoom Meeting
https://cern.zoom.us/j/66005783801?pwd=YW1mb2RrTFR5KzVMdWJjbFY1MVRZUT09
Meeting ID: 660 0578 3801
Passcode: 480101
