Public Lecture #2

"Mobile NPUs for Intelligent Human Computer Interaction"

Prof. Hoi-Jun YOO (KAIST)

Abstract

Deep learning is influencing not only the technology itself but also our everyday lives. In this presentation, the status of AI and DNN SoCs will be reviewed from the viewpoint of the mobile and edge AIs, and the evolution of DNN accelerators. Especially, low power NPU and reconfigurable NPU will be introduced for mobile, embedded, and IoT platforms. In addition, “Dynamically Reconfigurable Processor” architecture will be explained in detail with the real chip measurement results, such as human emotion recognition for intelligent HCI. Moreover, the on-chip training, which will be the next wave over the current AI revolution, and low power training processors will be explained with algorithms and hardware co-optimization methods for personalized and privacy protected AI applications. Real-time training, GAN and DRL examples will be introduced with dedicated training chips which are implemented for the many exciting low-power and high performance applications such as object recognition and the object tracking applications in Gaming, AR/VR, Intelligent Robotics, Drones, Autonomous Driving, Security, Camera, Health Monitoring and IoT. Lastly, fundamentals of neuromorphic architecture, which are realized with custom memories including non-volatile memories, will be introduced.

Back