Edge Intelligence based Co-training of CNN

Abstract

Improving training efficiency is a long-term major topic of neural network. With the popularity of intelligent terminal devices, such as smart phones and smart household electrical appliances, edge computing is gradually developing to edge intelligence (EI), which provides support to multiple terminals and play a role of a mini cloud. By making full use of the computing power of intelligent terminals in EI, while cooperating with edge servers, the training efficiency of convolution neural network (CNN) can effectively be improved. In this paper, an EI based co-training model between the edge server and intelligent terminals is proposed. The edge server and terminals propagate part of the CNN separately to accelerate the training of the CNN. Running time is greatly cut down by the division of CNN and simultaneous propagation of multiple terminals.