 |
    |
Shuffle and Mate: A Dynamic Model for Spatially Structured Evolutionary Algorithms
Carlos M. Fernandes1, 2, Juan L.J. Laredo3, Juan Julian Merelo2, Carlos Cotta4, Rafael Nogueras4, and Agostinho C. Rosa1
1LaSEEB-ISR-IST, University of Lisbon, Portugal
cfernandes@laseeb.org
acrosa@laseeb.org
2Department of Architecture and Computer Technology, Univ. of Granada, Spain
jjmerelo@gmail.com
3Faculty of Sciences, Technology and Communications, Univ. of Luxembourg, Luxembourg
juan.jimenez@uni.lu
4Departamento de Lenguages y Ciencias de la Computación, Univ. of Malaga, Spain
ccottap@lcc.uma.es
rafael.nogueras@gmail.com
Abstract. This paper studies a self-organized framework for modeling dynamic topologies in spatially structured Evolutionary Algorithms (EAs). The model consists of a 2-dimensional grid of nodes where the individuals interact and self-organize into clusters. During the search process, the individuals move through the grid, following a pre-defined simple rule. In order to evaluate the model, a dynamic cellular Genetic Algorithm (dcGA) is built over the proposed topology and four different movement rules are tested. The results show that when the ratio between the number of nodes in the grid and the population size is above 4:1, the individuals self-organize into highly dynamic clusters and significantly improve results attained by standard cGAs with static topologies on a set of deceptive and multimodal functions.
LNCS 8672, p. 50 ff.
lncs@springer.com
© Springer International Publishing Switzerland 2014