Accelerate Deep Learning in IoT: Human-Interaction Co-Inference Networking System for Edge

Abstract

As the core technology of the artificial intelligence in the new era, AI technology applied in health care devices has received significant attention. However, due to the limitation of the power supply and computation resource, it is difficult to implement a stable and large AI based human interaction task processing system from the remote edge devices to the centered clouds. In this paper, we propose a holistic network solution that focuses on solving the potential problems of network congestion with the explosive growth of IoT health care devices supported AI inference tasks. First, we propose a multi-hop maximum weight network to describe a DNN inference network based on edge computing. Then, we propose a Maximum Weight Wave propulsion Algorithm (MWWP) algorithm to reduce the overall network latency. Finally, we build up a prototype of a distributed AI inference system and test the computation and transmission performance. Besides, through large-scale experiments, we prove the optimality of our holistic solution.