Order statistics in artificial evolution

Stéphane Puechmorel (1) , Daniel Delahaye (2)

1 ENAC - Ecole Nationale de l'Aviation Civile
2 DGAC - Direction Générale de l'Aviation Civile

This article deals with the exploitation of statistical information from extremes values of an evolutionary algorithm. One can use the fact that upper order statistics of a sample converge to known distributions for improving efficiency of selection and crossover operators. The work presented in this paper is restricted to criteria defined on real vector spaces. It relies on an underlying canonical model of genetic algorithm, namely tournament selection and uniform crossover. Nevertheless, the results obtained so far encourage further investigations.

Mots clés: vector space; genetic algorithm; evolutionary algorithm; extreme value; probabilistic approach; statistical analysis; tournament; espace vectoriel; algorithme génétique; algorithme évolutionniste; valeur extrême; approche probabiliste; analyse statistique; tournoi

Domaines: Mathématiques [math]; Optimisation et contrôle [math.OC]

Stéphane Puechmorel, Daniel Delahaye. Order statistics in artificial evolution. Lecture Notes in Computer Science, 2003, 2936, pp 51-62. ⟨10.1007/978-3-540-24621-3_5⟩. ⟨hal-01004107⟩