Towards Independent On-device Artificial Intelligence
Yawen Wu, Jingtong Hu · 2022
By deploying AI to edge devices, on-device AI can power various tasks in our lives, from augmented reality (AR) / mixed reality (XR) glasses, daily life assistance in smartphones, health care in robots, to search and rescue in Unmanned Aerial Vehicles (UAVs). However, deploying DL to edge devices and applying them to real applications are challenging. The computational and energy cost of model inference is prohibitively large for edge devices with limited computation resources and battery capacity, which prevents real applications of on-device AI. Besides, the pre-trained models cannot dynamically adapt to the real world after being deployed to edge devices and may result in low accuracy for new input instances. To achieve efficient and adaptive on-device AI, two projects are carried out.