Real-time AI at the Edge: Inference, Kernels and Accelerators

Real-time AI at the Edge is a Tokyo AI (TAI) technical evening on making AI respond quickly when compute, memory and power are limited. It is aimed at ML engineers, systems developers and engineers building edge AI products, and connects low-level implementation choices with the latency and reliability users experience. Antoine Sauvage (EDGEMATRIX) shares lessons from building real-time computer vision applications on NVIDIA Jetson, from deploying an optimized model to running it reliably in the field with GPU-accelerated video pipelines, unreliable cameras and networks, and long-term maintenance. Xiao Wang (Renesas Electronics) explores using AI agents to coordinate hardware-aware optimization, reuse deployment knowledge and validate results. A third talk on GPU profiling and finding inference bottlenecks has its speaker still to be announced. The talks are followed by an hour of networking; the date is provisional and the Tokyo venue is to be announced.