Communication-efficient decentralized cooperative data analytics in sensor networks

Abstract

This paper presents a novel approach enabling communication-efficient decentralized data analytics in sensor networks. The proposed method aims to solve the decentralized consensus problem in a network such that all the nodes try to estimate the parameters of the global model and they should reach an agreement on the value of the model eventually. Our algorithm leverages broadcasting communication and is performed in a asynchronous manner in the sense that each node can update its estimate independent of others. All the nodes in the network can run the same algorithm in parallel and no synchronization is required. Numerical experiments demonstrate that the proposed algorithm outperforms the benchmark, and it is a promising approach for big data analytics in sensor networks.