Efficient Computer Morphogenesis: A Pictorial Demonstration

Abstract

This paper illustrates that artificial morphogenesis can be a computationally efficient technique. Artificial morphogenesis can develop graph grammar into modular Artificial Neural Networks (ANN), made of a combination of more simple sub-networks. The genetic algorithm is used to evolve coded grammar that generate ANNs for a simplified six-legged robot. The genetic algorithm can automatically decompose a problem into sub-problems, generate a sub-ANN for solving the sub-problem, and instanciate copies of this sub-ANN to build a higher level ANN that solves the problem. We support our argumentation with pictures describing the morphogenesis and illustrating how ANN structures are evolved.