摘要：AbstractMaximum consensus refers to all nodes in the network reach to the maximum value of the initial states through local communication. It has been applied in many domains, such as sensor network, social network, etc. When privacy is concerned, nodes’ states may contain some sensitive information, which nodes are willing to protect simultaneously. Thus, privacy-preserving maximum consensus (PPMC) algorithm has been proposed to solve this problem. However, PPMC doesn’t take eavesdroppers into consideration while differential privacy guarantees nodes’ states not to be disclosed to an adversary even looking at all transmitted states. In this paper, we first prove the existing PPMC algorithm cannot preserve ϵ-differential privacy. Then, we propose a differentially private maximum consensus (DPMC) algorithm, where nodes add Laplacian noises to initial states for communication. We prove DPMC algorithm is ϵ-differentially private while achieving certain convergence accuracy. The tradeoff between the convergence accuracy and privacy preserving performance is analyzed.
