A Reduced-Cost SMS-EMOA Using Kriging, Self-Adaptation, and Parallelization

The SMS-EMOA is a simple and powerful evolutionary metaheuristic for computing approximations to Pareto front based on the dominated hypervolume indicator (S-metric). However, as other state-of-the-art metaheuristics, it consumes a high number of function evaluations in order to compute accurate approximations. To reduce its total computational cost and response time for problems with time consuming evaluators, we suggest three adjustments: Step-size adaptation, Kriging metamodeling, and Steady-State Parallelization. We show that all these measures contribute to the acceleration of the SMS-EMOA on continuous benchmark problems as well as on a application problem – the quantum mechanical optimal control with shaped laser pulses.