法提赫·波里克利Fatih Porikli

Runs AI research at Qualcomm, where the constraint is not how large a model can be but how small it can get and still be worth running on a phone, a pair of glasses or a car: quantisation, distillation, efficient architectures and the compiler work that turns them into something a Hexagon NPU actually executes. Before Qualcomm he was a professor at ANU and a principal scientist at Mitsubishi Electric Research Labs, with a long computer-vision record in tracking and super-resolution. His practical argument is that on-device inference wins on latency, cost and privacy well before it wins on raw quality.