IRIS-HEP Fellow: Abdelrahman Elabd
Fellowship dates: Jan – Jul, 2021
Home Institution: University of Pennsylvania
Project: Creating a Python Front-End for HLS Implementation of GNNs on FPGAFPGA-implementation of GNN particle-tracking algorithms can reduce latency to the speeds necessary for tracking at the LHC, but the catch is that FPGA/HLS design is time and effort intensive. This hinders the speed with which we can implement and test new GNN structures and training paradigms. This project proposes to develop a Python front-end which converts trained pytorch.geometric GNN models into identical HLS representations, and to integrate this functionality into the hls4ml toolkit. This functionality will allow us to keep up with new developments in GNNs such as quantization-aware training, which may further reduce latency and resource usage while providing less lossiness than post-training quantization.
More information: My project proposal
Markus Atkinson (UIUC)
- 7 Jul 2021 - "Developing a Python Front-End for HLS Implementation of GNNs on FPGA and Studies of Quantization-aware Training", Abdelrahman Elabd, IRIS-HEP Topical Meetings Recording: Developing a Python Front-End for HLS Implementation of GNNs on FPGA and Studies of Quantization-aware Training
December 2021 - Data Scientist at TDI Technologies, Inc