Efficient On-Device Deep Learning: Challenges and Opportunities

Explore efficient on-device deep learning in this 25-minute conference talk from tinyML Asia 2021. Delve into the challenges and opportunities of edge computing as Yunxin Liu, Guoqiang Professor at Tsinghua University's Institute for AI Industry Research, discusses the shift from centralized cloud intelligence to distributed edge intelligence. Discover recent research on customizing affordable AI models for diverse edge devices and maximizing on-device model inference performance through heterogeneous computing resources. Learn about hardware-friendly neural networks, kernel-level latency prediction, mobile CPU utilization, and optimization techniques. Gain insights into the future of AI-empowered edge devices and applications in this comprehensive overview of on-device deep learning advancements.

Introduction
Ideal computing
Data explosion
Four directions
Two directions
Hardware friendly neural networks
Kernel level latency prediction
Mobile CPU utilization
Optimization
Results