IRIS-HEP Fellow: Aneesh Heintz
Fellowship dates: Jun – Sep, 2020
Home Institution: Cornell University
Project: Implementation of graph neural networks on CPU + FPGA co-processors for scalable track reconstruction tasksRun 2 of the LHC on the CMS detector produced a data rate on the scale of hundreds of terabytes per second. Being able to reduce the data within a few milliseconds and sift through the data in a reasonable time frame to produce meaningful results is crucially important. Future increases in instantaneous luminosity, meaning more proton-proton collisions per bunch-crossing, will lead to data produced at increasingly larger rates, causing scalability issues in traditional particle track reconstruction algorithms. This project proposes to implement a graph network that can be evaluated on a CPU that has a FPGA co-processors. This will allow trained networks to be run online in a highly parallelized fashion, greatly accelerating data throughput.
More information: My project proposal
Isobel Ojalvo (Princeton University)
- 5 Oct 2020 - "Accelerating Graph Neural Networks on CPU + FPGA co-processors for scalable track reconstruction tasks", Aneesh Heintz, IRIS-HEP Topical Meetings Recording: Accelerating Graph Neural Networks on CPU + FPGA co-processors for scalable track reconstruction tasks
July 2022 - Senior Autonomy / AI Research Engineer at Lockheed Martin