Property | Value |
Name | NCL-TR-2007008 |
Description |
Linkage Identification by Perturbation and Decision Tree Induction
Abstract: The purpose of linkage identification in genetic and evolutionary algorithms is to detect the tightly linked variables of the fitness function. If such linkage information is not known a priori and can be obtained through computation, the crossover or recombination operator can accordingly mix discovered sub-solutions effectively without disrupting them. In this paper, we propose a linkage identification technique, called inductive linkage identification (ILI), employing perturbation with decision making. With the proposed scheme, the linkage information can obtained by constructing an ID3 decision tree to learn the mapping from population of solutions to their corresponding fitness differences caused by perturbation and inspecting the constructed decision tree for variables exhibiting strong interdependencies with one another. The numerical results show that the proposed technique can accomplish the identical linkage identification tasks with a lower number of function evaluations compared to similar methods proposed in the literature. Moreover, the proposed technique is shown able to handle both uniformly scaled and exponentially scaled problems. |
Filename | NCL-TR-2007008.pdf |
Filesize | 198.31 kB |
Filetype | pdf (Mime Type: application/pdf) |
Creator | ypchen |
Created On: | 04/02/2007 10:30 |
Viewers | Everybody |
Maintained by | Publisher |
Hits | 3050 Hits |
Last updated on | 12/09/2010 14:26 |
Homepage |