next up previous contents
Next: 2 Code Growth and Up: 3 Future Directions Previous: 3 Future Directions   Contents

1 Diversity Measures and Methods

Developing measures based on the operators would give an improved understanding of diversity and of the fitness landscape. However, the complexity of the representation would make these measures computationally expensive. Approximate measures based on the operators would require careful consideration of their accuracy. Intermediate data structures could be used improve the efficiency of such complex measures.

Chapter 5 used a new diversity method based on genetic lineages. Additional research with this method could yield more efficient ways to control diversity and search, or easily simulate mating, niching and island models. Chapter 6 suggested controlling diversity by adapting selection pressure, which could lead to slower and more controlled code growth without fewer negative effects to fitness. Future work could also characterise the ``effective'' selection pressure that populations with varying diversity induce.

S Gustafson 2004-05-20