Optimization with auxiliary criteria using evolutionary algorithms and reinforcement learning

In this paper, an optimization method EA + RL based on an evolutionary algorithm controlled by reinforcement learning is proposed. Reinforcement learning is used to choose the most effective fitness function at each generation of the evolutionary algorithm. The method can be applied in scalar optimization with auxiliary criteria to speed up the optimization process. Experimental results for a model problem H-IFF are given. Applying of the method doubles mean fitness obtained with evolution strategy. A comparison with other evolutionary optimization methods is performed. The proposed method outperforms all the considered scalar optimization methods and most of the multicriteria ones.