The Smart Cities Research Institute (SCRI) successfully hosted a Distinguished Lecture on 8 July 2025, focusing on the cutting-edge field of "Domain-Specific Acceleration Targeting AI for Science". The event featured the internationally renowned Professor Wayne Luk from Imperial College London, UK, drawing an engaged audience of over 60 participants. The lecture explored the critical role of domain-specific acceleration in advancing computer system design. He emphasized that tailoring hardware accelerators to specific computational tasks is fundamental to unlocking the full potential of Artificial Intelligence (AI), particularly for demanding scientific workloads. This customisation enables the adoption of sophisticated AI techniques across diverse application domains. Illustrating this concept with groundbreaking work, Professor Luk detailed his team's collaboration with high-energy physics researchers at CERN. He presented the development of ultra-low-latency deep learning accelerator architectures designed for the extreme conditions of the CERN Large Hadron Collider (LHC) experiments. A highlight was the introduction of a bespoke Graph Neural Network (GNN) accelerator capable of achieving sub-microsecond latency. This breakthrough performance is crucial for real-time event selection within the LHC's Level-1 trigger systems, where decisions must be made in billionths of a second. Furthermore, Professor Luk discussed ongoing efforts to streamline the creation of such specialized accelerators using meta-programming techniques, aiming to automate and expedite the development process for future scientific AI applications. The lecture concluded with a dynamic Q&A session, fostering insightful discussion between attendees and Professor Luk on the challenges and future directions of domain-specific acceleration for scientific AI.
| Research Units | Otto Poon Charitable Foundation Smart Cities Research Institute |
|---|
