Hierarchical Breeding Control for Efficient Topology/Parameter Evolution

Kisung Seo^1,2, Jianjun Hu¹, Zhun Fan¹, Erik D. Goodman¹, and Ronald C. Rosenberg¹

¹Genetic Algorithms Research and Applications Group (GARAGe), Michigan State University, East Lansing, MI 48824, USA
ksseo@skuniv.ac.kr
hujianju@egr.msu.edu
fanzhun@egr.msu.edu
goodman@egr.msu.edu
rosenber@egr.msu.edu

²Department of Electronics Engineering, Seokyeong University, Seoul, 136-749, Korea

Abstract. This paper adopts a hierarchical breeding control mechanism to obtain better search performance based on differential balancing of topology-altering operations and parameter- altering operations according to fitness level, in a fitness-structured multipopulation model. The basic idea for this control mechanism arises from observing the human design process. Usually, preliminary or conceptual design involves more structural modification, and final or detailed design involves more parameter tuning – i.e., there is greater concentration on design topology in the early stage and more on parameter tuning in the later stage. Therefore, the key concept is to provide different breeding probabilities for topology-altering and parameter-altering operations according to fitness level of the subpopulation.

LNCS 3103, p. 722 f.

Full article in PDF

lncs@springer.de
© Springer-Verlag Berlin Heidelberg 2004