Investigating surrogate-assisted cooperative coevolution for large-Scale global optimization

Problems involving large-scale global optimization (LSGO) are becoming more and more frequent. For this reason, the last few years have seen an increasing number of researchers interested in improving optimization metaheuristics in such a way as to cope effectively with high-dimensional search domains. Among the techniques to enhance scalability, one of the most studied is Cooperative Coevolution (CC), an effective divide-and-conquer strategy for decomposing a large-scale problem into lower-dimensional subcomponents. However, despite the progress made in the LSGO field, one of such optimizations can still require a very high number of objective function evaluations. Therefore, when the evaluation of a candidate solution requires complex calculations, LSGO can become a challenging task. Nonetheless, to date few studies have investigated the application of optimization metaheuristics to objective functions that are simultaneously high-dimensional and computationally significant. To address such a research issue, this article investigates a surrogate-assisted CC (SACC) optimizer, in which fitness surrogates are exploited within the low-dimensional subcomponents resulting from the problem decomposition. The SACC algorithm is investigated on a rich test-bed composed of 1000-dimensional problems. According to the results, SACC is able to significantly boost the convergence of the CC optimizer, leading in many cases to a relevant computational gain.