Dynamics Based Privacy Protection for Average Consensus on Directed Graphs (1812.02255v1)

Abstract: Average consensus is key for distributed networks, with applications ranging from network synchronization, distributed information fusion, decentralized control, to load balancing for parallel processors. Existing average consensus algorithms require each node to exchange explicit state values with its neighbors, which results in the undesirable disclosure of sensitive state information. In this paper, we propose a novel average consensus approach for directed graphs which can protect the privacy of participating nodes' initial states without the assistance of any trusted third party or data aggregator. By leveraging the inherent robustness of consensus dynamics to embed privacy in random coupling weights between interacting nodes, our proposed approach can guarantee consensus to the exact value without any error. This is in distinct difference from differential-privacy based average consensus approaches which enable privacy through sacrificing accuracy in obtained consensus value. The proposed approach is able to preserve privacy even when multiple honest-but-curious nodes collude with each other. Furthermore, by encrypting exchanged information, the proposed approach can also provide privacy protection against inference by external eavesdroppers wiretapping communication links. Numerical simulations and hardware experiments on Raspberry Pi boards confirm that the algorithm is lightweight in computation and communication.