|Fitness Landscape for Simple Genetic Algorithms Supplied with Adequate Superior Order-1 Building Blocks
Hongqiang Mo, Zhong Li*, Jin Bae Park, Young Hoon Joo, and Xiangyang Li
International Journal of Control, Automation, and Systems, vol. 8, no. 1, pp.135-140, 2010
Abstract : Building block hypothesis suggests that the highly-fit low-order schemata recombine with each other to form even more highly-fit high-order ones. One may naturally surmise that the coding should be designed to supply adequate superior order-1 schemata. In this paper, it is showed that, if superior order-1 building blocks are provided at most of the loci, there is likely to be remarkable fitness differences among high-order schemata, which indicates the existence of ‘pulse-shaped’ peaks on the curve of the fitness function. And fitness differences among the individuals are so great within the neighborhoods of these peaks that diversity loss tends to occur when searching within these regions. The results of this paper may to some degree explain why additional measures to maintain diversity should be taken to improve the local search performance of a simple genetic algorithm (GA).
Keyword : Building block, diversity loss, fitness landscape, parallel search, simple genetic algorithms.