On sampling methods for costly multi-objective black-box optimization

Abstract

We investigate the impact of different sampling techniques on the performance of multi-objective optimization methods applied to costly black-box optimization problems. Such problems are often solved using an algorithm in which a surrogate model approximates the true objective function and provides predicted objective values at a lower cost. As the surrogate model is based on evaluations of a small number of points, the quality of the initial sample can have a great impact on the overall effectiveness of the optimization. In this study, we demonstrate how various sampling techniques affect the results of applying different optimization algorithms to a set of benchmark problems. Additionally, some recommendations on usage of sampling methods are provided.