Coevolutionary strategies at the collective level for improved generalism

Abstract In many complex practical optimization cases, the dominant characteristics of problem are often not known prior. Therefore, there is a need to develop general solvers as it always possible tailor specialized approach each application. The previously developed multilevel selection genetic algorithm (MLSGA) already shows good performance on range problems due its diversity-first approach, which rare among evolutionary algorithms. To increase generality performance, this paper proposes utilization multiple distinct strategies simultaneously, similarly selection, but with coevolutionary mechanisms between subpopulations. This distinctive coevolution provides less regular communication subpopulations competition collectives rather than individuals. encourages act more independently creating unique subregional search, leading development MLSGA (cMLSGA). test methodology, nine algorithms selected generate several variants cMLSGA, incorporates these approaches at individual level. tested 100 different functions and benchmarked against 9 state-of-the-art competitors evaluate approach. results show that diversity divergence in principles working important their performances. proposed methodology has most uniform divergent types, from across state art, an likely solve limited knowledge about search space, outperformed by simpler benchmarking studies.