Privacy-Preserved Distributed Optimization for Multi-Agent Systems With Antagonistic Interactions

This paper is concerned with the privacy-preserving distributed optimization problem for a class of cooperative-competitive multi-agent systems. Each agent only knows its own local objective function and interacts the state information with neighbors through a communication network. By means of the signed graph theory, the antagonistic interactions among agents are considered to characterize both the cooperative and the competitive relationships. With the help of the gauge transformation technique, a structurally balanced undirected signed graph is firstly transformed into a standard undirected graph. Then, the distributed optimization problem subject to signed network is converted into the traditional distributed optimization problem. Subsequently, a novel privacy-preserving distributed optimization algorithm is put forward to 1) minimize the sum of local objective functions; 2) achieve the bipartite consensus for all agents; and 3) avoid the information leakage caused by message exchange among agents, simultaneously. Finally, a simulation example is given to verify the effectiveness of the proposed optimization algorithm.

Automated summary

This paper addresses the privacy-preserving distributed optimization problem for a class of cooperative-competitive multi-agent systems. By utilizing signed graph theory and gauge transformation technique, a novel privacy-preserving distributed optimization algorithm is proposed to tackle the issues of information leakage and cooperative-competitive relationships.