A Taxonomy of the Evolution of Artificial Neural Systems
Helmut A. Mayer · 2005
Abstract. Biological neural networks having been shaped in billions of years of evolution are the source of numerous, complex capabilities of living organisms. Especially, the form of intelligence attributed to humans has inspired computer scientists to model the evolution of neural networks with the ultimate goal to create computer systems with cognitive abilities. However, considering the cur-rent state of research in the domain of artificial neural network (ANN) evolution there is still a huge gap between the extremely versatile brains of higher organisms, and the very specialized artificial neural systems mostly applied to a single, well–defined task. The majority of work on ANN evolution is concerned with the optimization of network structure and/or network weights so as to achieve maximal network performance for a specific problem. We present a taxonomy in order to categorize the (co)evolution of various components of an artificial neural system (ANS). We illustrate some of the approaches in ANS evolution by a few examples from the literature, and present our extensions to evolution of the ANN learning component by means of experiments with the two spirals benchmark problem utilizing the netGEN system. Specifically, we demonstrate that the evolution of non–monotonous activation functions of (nearly) arbitrary shape can enhance the performance of an ANS. 1.