An RTS-based algorithm for noisy optimization by strategic sample accumulation

This paper proposes a noisy optimization algorithm that can effectively and efficiently deal with the noise in the objective function by appropriately allotting samples to individuals in the population of an overlapping generation model. During the survival selection, our algorithm makes probabilistic decisions to determine when and to whom additional samples should be given, intending to maximally save the samples. Since the samples allotted to individuals are accumulated as long as they remain in the population, long-lasting individuals tend to have a large accumulation and thus a quite accurate evaluation. Test results with benchmark functions confirm the performance of the proposed algorithm.