Technical Seminar by Professor Amin Arbabian from Stanford University with the following abstract: AI-driven machine perception is transforming domains such as robotics, healthcare, consumer electronics, and the IoE. As neural networks scale and edge sensors generate ever-larger data volumes, inference tasks are becoming increasingly resource-intensive, pushing up against computational limits. This talk explores two key aspects of this trend. First, we introduce a new 3D sensing paradigm that exemplifies the dramatic increase in data rates for next-generation physical AI systems. Second, we present a neuroscience-inspired adaptive inference framework that addresses processing bottlenecks in edge-based AI. We derive theoretical bounds and provide empirical results showing 10–100× efficiency gains in vision and language tasks. We further highlight how optimal design of adaptive inference state spaces can unlock even greater computational savings. Speaker(s): Amin Arbabian, Room: 3038, Bldg: Macleod Building, 2356 Main Mall, Vancouver, British Columbia, Canada, V6T 1Z4