Clustering-Based Edge Compression Method with Application to Electromagnetic Object Recognition

Collaborative intelligence has attracted more and more attention. By properly portioning deep neural networks (DNN) and distributing the DNN calculation to the edge and cloud, we could reduce the prediction delay and power consumption to meet the actual application requirements. Toward this direction, this paper proposes an edge compression method based on clustering to address the issue of high data communication cost and time delay between the edge and cloud. Specifically, by using K-means clustering algorithm, this method compresses the output layer of the edge DNN, reducing the amount of transmission data and thus the delay and energy consumption. Based on the compression-based edge-cloud collaboration paradigm, we propose a distributed inference scheme for electromagnetic object recognition. The simulation results show that the proposed method can greatly reduce communication cost while maintaining the prediction performance.