Task 1.2: Fast inference of complex network architectures on LHC online systems

Task leads: Vladimir Loncar (lead), Sebastian Ditmeier (deputy), Maurizio Pierini (deputy)
In this task, we work with existing expertise in the experiment collaboration on ongoing work on tools such as hls4ml, and on expertise from selected academic and industrial partners to develop ML->FPGA model synthesis tools, addressing the needs of WP2 and WP3. The work will also focus on integrating modern ML tooling while maintaining the strict latency requirements set forth by LHC experiments’ online selection system. All task items are supposed to be co-developed by CERN researchers and external partners with qualified expertise on the topic.
Publications and other resources
- Vladimir Loncar, Dimitrios Danopoulos, 08/08/2025, “Enhancing the FPGA synthesis service platform “Gofer””
- Dimitrios Danopoulos, Presentation at FastML 02/09/2025, “Pushing Matrix-Vector Multiplication Performance on AMD AI Engines for Low-Latency Edge Inference”
- Dimitrios Danopoulos, Presentation at ACAT, 09/09/2025, “Accelerating Deployment of FPGA-based AI in hls4ml with Parallel Synthesis through Model Partitioning”