Genetic programming is an appealing search heuristic because it provides an intuitive way to search a very complex space. This representation provides many possible measures and methods of diversity. It is not clear what type of diversity is more relevant for improving search, or what level of diversity would be best. Next, Chapter 4 looks at diversity measures and the correlation between fitness and diversity. Chapter 5 then uses a novel diversity measure to explore the effects of increased diversity, the expected diversity loss and the type of search genetic programming appears to be carrying out.
Issues such as code growth, bloat and problem difficulty and homology are not the primary focus of this thesis. Instead, Chapter 6 examines population diversity to understand its role and the effects it has on the previous issues. As the population is connected to most issues in search, it is likely that many interesting relationships exist.
The population and operator are responsible for producing new individuals during search. While diversity is obviously beneficial to sampling new individuals, do the most dissimilar individuals produce good individuals? Should the search method focus explicitly on dissimilar individuals or simply use them to provide variety for a converging population? These issues are fundamental to how genetic programming searches and also to developing new methods that can be used to improve search. Chapter 7 explores the concept of the survivability of dissimilar and fit individuals to better understand how these individuals guide search.