- List of Figures
- List of Tables
- 1 An Analysis of Diversity

- 2 Search, Evolutionary Algorithms and Genetic Programming
- 1 Problem Solving and Search
- 2 Evolutionary Algorithms
- 3 Genetic Programming
- 4 Application Domains
- 5 Scalability and Fitness Landscape
- 6 Metaphors of Search
- 7 Summary

- 3 Issues in Genetic Programming
- 1 Diversity Measures and Methods
- 2 The Effects of Population Diversity
- 3 The Role of the Population
- 4 Summary

- 4 Analysis of Diversity Measures
- 1 Diversity Measures
- 2 Empirical Analysis of Diversity Measures
- 3 Analysis of Results
- 4 Discussion of Diversity Measures
- 5 Summary

- 5 Genetic Lineages and A Metaphor of Hill Climbing
- 1 Genetic Lineages
- 2 Experimental Study using Lineage Selection
- 3 Results of Lineage Selection
- 4 Discussion of the Metaphor of Hill Climbing
- 5 Sampling of Unique Structures and Behaviours
- 6 Analysis of Results
- 7 Discussion of Sampling
- 8 Summary

- 6 Effects of Population Diversity: Code Growth and Problem Difficulty
- 1 Code Growth and Problem Difficulty
- 2 Regression Problems and Increased Difficulty
- 3 Experimental Investigation
- 4 Binomial-3 and Random Polynomial Results
- 5 Discussion of a Causal Model
- 6 Summary

- 7 Diversity, Survivability and a Niche for Island Models
- 1 Previous Distributed Evolution Work
- 2 Survivability of the Diverse
- 3 Genetic Outliers and Survivability
- 4 The Ant, Parity and Regression Domains
- 5 A Niche for Island Models in Genetic Programming
- 6 Summary

- 8 Conclusions
- 1 Contributions
- 1 A survey and analysis of diversity in genetic programming demonstrated the complexity behind the issue of diversity measures and methods and the relationship between diversity and fitness.
- 2 An analysis using genetic lineages showed how a search metaphor of hill-climbing can be used to explain and improve genetic programming search. Also, the sampling of unique structures and behaviours demonstrated the the low sampling of both complex behaviours and unique structures of large size.
- 3 A causal model was developed which linked increased rates of code growth to non-decreased selection pressure and to increased similarity within the population. Decreased selection pressure occurs when fitness-based diversity is lost, and increased similarity in the population is the result of both faster convergence and non-decreased selection pressure.
- 4 An analysis using the Tree-String problem showed the inability to produce good offspring by both dissimilar-and-fit individuals and by similar-and-equally-well-fit individuals.
- 5 A model was proposed that identifies dissimilar individuals and moves them to new islands where they can contribute to search more effectively.
- 6 Summary

- 2 Remarks and Problem Specific Conclusions
- 3 Future Directions

- 1 Contributions
- Bibliography
- About this document ...

S Gustafson 2004-05-20