Variance Reduction in Meta-EDA

Proceedings of the Genetic and Evolutionary Computation Conference (GECCO)

Category: Conferences Publication Date: December, 2007 Editor(s): Dirk Thierens and others Pages: 627 Citekey: Nannen-Eiben:2007b
Keywords: Estimation of Density Algorithms, parameter tuning, parameter control

We study the benefit of measurement replication when using the Relevance Estimation and Value Calibration method to calibrate a genetic algorithm. We find that replication is not essential to REVAC, which makes it a strong alternative to existing statistical tools which are computationally costly.