Empirical Evaluation of Evolutionary Algorithms with Power-Law Ranking Selection

The chapter focuses on the empirical evaluation of evolutionary algorithms with power-law ranking selection, a mechanism that assigns selection probabilities based on the ranks of individuals. It discusses the advantages of this selection method, such as its ability to handle small but important advantages in large populations and its compliance with conditions that prevent non-elitist EAs from getting stuck in local optima. The study presents extensive experimental results on benchmark functions, including NK-Landscape, MaxSat problems, and the combinatorial problem Set Cover. The experiments demonstrate that power-law selection allows non-elitist EAs to operate with high mutation rates and can outperform other selection mechanisms on certain problems. The chapter also provides a detailed analysis of the error threshold, which balances selection and mutation, and shows how power-law selection can improve the performance of evolutionary algorithms on complex optimization tasks.