2 Similar Models

A similar model was proposed by Bessaou et al. (2000) for multimodal optimisation with genetic algorithms. In their model, a multipopulation algorithm performed migration between subpopulations. The subpopulations were then merged and individuals were redistributed to new subpopulations based on a speciation tree method that places genetically similar individuals together. A local genetic algorithm was then run on each island for a number of generations to serve as an `intensification' phase. Our proposed model does not consider migration or the concept of merging the subpopulations before redistributing them according to species. The genetic algorithm species conserving model [Li et al., 2002] identified diverse individuals and assigned the fitter to be seeds which are propagated to the next generation. However, recombining diverse individuals, as seen in Section 7.3, may not always be ideal for genetic programming.

Fitness sharing over structures was used to balance the search between structure and contents [Hu et al., 2002]. However, like other fitness sharing methods in genetic programming based on genetic dissimilarity, this approach ignores the possible benefits of homology and the possible negative effects of recombination using diverse parents. Lastly, the method by Potter and DeJong (2000) for coevolution of components requires each species to contribute toward fitness to survive and new species are added when no improvements are found. In a similar way, future research with the proposed model can investigate different ways to remove stagnate islands while still allowing the possibility of variation to arise during stasis.