6 Effects of Population Diversity: Code Growth and Problem Difficulty

The issues of code growth and problem difficulty were introduced in Chapter 3. In Chapter 4, various forms of diversity were shown to have a complex relationship with fitness improvement, where edit distance and fitness-based measures seem to represent important properties. Chapter 5 furthered the analysis of diversity by illustrating expected diversity loss using the concept of genetic lineages and a caricature of tournament selection. Also, Chapter 5 highlighted a potential metaphor of genetic programming search and the signs and consequences of deception in the search space. How do these results apply to important issues such as code growth and problem difficulty?

In this chapter, diversity is closely examined with respect to problem instances that are both tunably difficult and that exhibit varying rates of code growth. The results strongly support a causal hypothesis relating code growth and diversity. This hypothesis is also supported by previous literature and further experiments using a constructed model of code growth and problem difficulty. While furthering the understanding of diversity and problem difficulty, this chapter also provides important insights that suggest new ways of addressing the issue of bloat.