| The culmination of this dissertation is to emulate nature's design strategy in the evolution of neural networks. Specifically we postulate that in order to capture the effectiveness of this mechanism (and avoid the commonly reported pitfalls or limitations), biologically inspired deviations from the general practice for evolving neural networks must be incorporated into the system. These system advancements include a more biologically realistic neural unit, a developmental genotype to phenotype translation paralleling neurogenesis, isolation and migration between populations, inclusion of regulatory genes in the genotype, and a design for modularity in functionality. By inclusion of these, we obtain a limitless potential for scalability of functionality of the resulting system, open-ended evolution, such as observed in the emergent phenomena of the human brain. The system and resulting simulations from each of its major functions are presented in this research work. Additionally, extensive investigations of the suitability of the proposed model in terms of evolvability of the system are recorded, concluding the effectiveness of the system design in obtaining its objective. |
