On the Privacy Issue of Evolutionary Biparty Multiobjective Optimization

Some evolutionary algorithms have been proposed to address biparty multiobjective optimization problems (BPMOPs). However, all these algorithms are centralized algorithms which directly obtain the privacy information including objective functions from decision makers (DMs). This paper transforms the centralized algorithm OptMPNDS2 into a distributed framework for BPMOPs and focuses on the privacy issue in the framework. The framework has a server and two clients, and each client belongs to a DM. The clients keep their objective functions locally, evaluate individuals, and upload Pareto levels and crowding distances of all individuals to the server. The server performs the other operations including reproduction and selection of offspring. Experimental results show that the performance of the framework is very close to OptMPNDS2. Besides, two privacy attacks are proposed when one client is malicious. Experimental results show that the client could steal approximate Pareto optimal solutions of the other honest DM.