Evolutionary algorithms and cross entropy

Article type: Research Article

Authors: Fyfe, Colin^{a; *} | Ortiz-Boyer, Domingo^b | Garcia-Pedrajas, Nicolas^b

Affiliations: [a] The University of the West of Scotland, Paisley, UK | [b] Cordoba University, Cordoba, Spain

Correspondence: [*] Corresponding author: Colin Fyfe, The University of the West of Scotland, Paisley, UK. E-mail: [email protected]

Abstract: Cross entropy is a method designed to estimate some statistic pertaining to events of very low probability. We discuss cross entropy with respect to optimisation problems and then illustrate the cross entropy method on a specific function (Rosenbrock's function) which we have found to be difficult to optimise using evolutionary algorithms. We examine the convergence of the cross entropy method to identify why evolutionary algorithms find this difficult. We then use a concept from evolutionary algorithms (that of separate sub-populations) to enhance the cross entropy method.

Keywords: Function optimization, convergence

DOI: 10.3233/KES-2012-00244

Journal: International Journal of Knowledge-based and Intelligent Engineering Systems, vol. 16, no. 4, pp. 215-221, 2012