Online Learning for Hierarchical Inference

Learn about cutting-edge developments in Hierarchical Inference (HI) through this technical talk by Prof. Sharayu Moharir from IIT Bombay's Electrical Engineering Department. Explore how HI enables efficient distributed inference between end devices with small pre-trained Deep Learning models and edge/cloud servers running larger models. Discover novel algorithms with sub-linear regret for optimizing confidence metric thresholds in HI systems, demonstrating significantly reduced computational complexity compared to existing approaches. Examine practical implementation results from Raspberry Pi runtime measurements that show improved performance while maintaining comparable cumulative loss metrics. Gain insights from Prof. Moharir's extensive research background spanning TIFR Mumbai, University of Texas at Austin, and IIT Bombay as she presents this advancement in distributed machine learning systems.