Publication at Faculty of Mathematics and Physics |
2012
Abstract
Estimation of distribution algorithms (EDAs) have become promising kinds of evolutionary algorithms. They are based on sampling an estimated distribution of the better of the solutions from the last generation -- iteration of the algorithm.
Learning the distribution and sampling is used instead of reproduction common in traditional evolutionary algorithms. In many real-world optimization tasks, objective-function-evaluation of any solution is very expensive, so the lowest possible number of such evaluations is desired.
We propose using surrogate models in combination with EDAs as a method of reducing the evaluation costs.