Authors:
Marcin Komarnicki
and
Michal Przewozniczek
Affiliation:
Wroclaw University of Science and Technology, Poland
Keyword(s):
Island Models, Coevolution, Genetic Algorithms, Linkage Learning, Messy Coding.
Related
Ontology
Subjects/Areas/Topics:
Artificial Intelligence
;
Co-Evolution and Collective Behavior
;
Computational Intelligence
;
Evolutionary Computing
;
Genetic Algorithms
;
Informatics in Control, Automation and Robotics
;
Intelligent Control Systems and Optimization
;
Soft Computing
Abstract:
Island Models (IMs) divide the whole population into many coevolving subpopulations, which periodically exchange fractions of their individuals. Some IMs, exchange probabilistic models built during the subpopulations evolution. The use of many coevolving subpopulations helps to preserve the population diversity, which makes it less likely to get stuck in the local optima. Another promising research direction in the Evolutionary Computation field is the Linkage Learning. The knowledge about gene dependencies can be used in many different ways that improve the overall method effectiveness. Therefore, this paper proposes the Gene Pattern Based Island Model (GePIM) that uses the multi-population nature of IMs to generate the linkage information. GePIM also introduces a new type of migration based on exchanging linked gene groups, instead of exchanging the whole individuals or probabilistic models.