IRIS-HEP Fellow: Farouk Mokhtar
Fellowship dates: Apr – Sep, 2021
Home Institution: University of California, San Diego
Project: GPU-accelerated Machine-learned Particle-flow ReconstructionAn important core software algorithm of the LHC is the particle-flow (PF) reconstruction algorithm, which takes disparate types of tracks and clusters reconstructed from different subdetectors and returns a list of final-state PF candidates. The nature of this task motivates the exploration of highly parallelizable machine learning (ML) models that are easier to accelerate with heterogeneous computing resources, such as GPUs and FPGAs, which gives them an advantage over traditional PF algorithms. This project proposes to apply state-of-the-art ML techniques, mainly graph neural networks (GNNs), which learn from non-Euclidean structured data, to the task of PF reconstruction in CMS and for LHC detectors more generally. Concrete deliverables of the project include providing publicly-available ML models for PF reconstruction and benchmarking their physics and computational performance on open datasets with coprocessing accelerators.
More information: My project proposal
Javier Duarte (University of California, San Diego)
- 20 Sep 2021 - "Machine-Learned Particle Flow", Farouk Mokhtar, IRIS-HEP Topical Meetings Recording: Machine-Learned Particle Flow
December 2021 - PhD Candidate in Experimental High Energy Physics University of California San Diego