3 Defining the Role of the Populations

This thesis has provided a better understanding of the role of the genetic programming population. However, several areas of the research in Chapter 7 could be further examined. For example, the exact dissimilarity between mating parents, the structure of dissimilarity in the un-fit population and in-depth analysis of survivability and genetic outliers in other domains are potential areas of future research.

If genetic programming is considered as a stochastic hill-climber using a procedural representation, the role of the population is to provide genetic material in a more elitist algorithm. If genetic programming is to use the population to sufficiently explore the landscape in parallel, one needs to explicitly maintain portions of the population in beneficial areas and dispense sufficient search effort appropriately. The proposed niche for island models is an obvious extension to the research presented here and gives the population a much clearer role during search. Empirical studies, some of which were described in detail in Chapter 7, need to be carried out to validate the proposed island model.

Chapter 7 also introduced the Tree-String problem. Investigating the tunable nature of this problem would validate it as a future testbed for understanding complex domains. Future work can begin by modifying the size of the string symbol set and the method to grow target tree structures.