next up previous contents
Next: 1 Evolutionary and Genetic Up: 7 Diversity, Survivability and Previous: 7 Diversity, Survivability and   Contents

1 Previous Distributed Evolution Work

Many methods have been proposed and used in evolutionary computation to control diversity, prevent convergence and to distribute individuals over different areas of the search space. These include niching methods, diversity methods, mate selection techniques and distributed population models.

The island model [Cohoon et al., 1987] is an example of a distributed population model where subpopulations are isolated during selection, breeding and evaluation. Islands focus the evolutionary process within subpopulations before migrating individuals to other islands. There are many variations of distributed models, e.g. islands, demes, and niching methods. However, by tuning the parameter settings of each model, the functionality of different models can become very similar. So, the different parameter and algorithm settings largely define the objectives of these models with respect to similarity and dissimilarity mating, elitism and selection pressure.

Other methods simulate distributed evolution by using specific forms of selection and recombination. For example, the method of lineage selection in Chapter 5 could be implemented to only allow the recombination of individuals from the same lineage, which are very likely to have similar shape and content. In effect, lineage selection would encourage a kind of similarity-based mate selection. Previous methods that focus selection and recombination on particular individuals, or on particular features of individuals, are also included in this section. Initially, methods from evolutionary algorithms are examined, followed by applications specific to genetic programming and the tree representation.



Subsections
next up previous contents
Next: 1 Evolutionary and Genetic Up: 7 Diversity, Survivability and Previous: 7 Diversity, Survivability and   Contents
S Gustafson 2004-05-20