摘要: Biological development is highly complex, beginning with an egg and resulting in a complete, living organism (Purves and Lichtman, 1985). Development is essentially sequential, establishing a gross structure which becomes progressively more complex over time (Goodwin, 1991). This refinement of structure and function/behaviour operates across many different levels of the biological scale, from molecules to cells to tissues and organs. On each level of scale there is interactive self-organization between the constituent elements (Goodwin, 1996). Neural development is an example of these processes which leads to the development of a nervous system and associated functions. For an abstract view of neural development see Figure 19.1. Modelling the nervous system as computational artificial neural networks has long been the source of interest to engineers and computer scientists (McCulloch and Pitts, 1943). Mathematical models have been widely used to model the functions and behaviour of the nervous system and, from an engineering perspective, to make use of the powerful, parallel nature of the nervous system to solve complex problems. However, few models have explored the potential of using neural development to automate the process of designing artificial networks, bypassing the need for hand-coding the initial architectures. But why should biological development be of interest to computer scientists and engineers who design artificial neural networks/systems? From a systems perspective, biological neural development encompasses many powerful ideas including:Variation. Development automatically produces a huge …