New multi-objective optimization strategy mechanism and its application
一种新的多目标优化策略机制及其应用

There are many learning mechanisms on game problems. Although most of them can make Agent convergence to Nash equilibrium solution, they can't satisfy the actual requirement. In this paper, we translate the game problems into multi-objective optimization problems, and present a new multi-objective optimization strategy mechanism-Reserve the Dominated Strategies (RDS). Empirical result proves that the Agent using this mechanism in self-play gets more significative Pareto Optimum Solution than Nash Equilibrium. The result of the experiment indicates the validity of the strategy to solve Pareto Optimum Solution.