Stochastic algorithms, differential evolution, measures of performance, statistical tests, comparison of experimental results

Problem of one-objective global optimization with boundary constraints and principles of evolutionary algorithms for its solution are presented. Measures for assessment of algorithm's efficiency are discussed. Basic terms from probability and mathematical statistics are briefly recapitulated. Non-parametric tests frequently used in comparison of algorithms, namely Wilcoxon's signed-rank and rank-sum tests are explained in more details. Consequences of improper application of paired test instead of two-sample test are illustrated on the results of a simulation experiment. Some advanced statistical methods for comparison of algorithms are also mentioned.