ABSTRACT
This chapter aims to simulate a variable population of network automata. In a manner directly analogous to biological evolution, the population converges, under the influence of selective pressure, to a group of networks that are well suited for solving the task at hand. It considers two separate kinds of processes: a low-level "performance" process that operates on a population of "phenotypes," and a higher level "metaperformance" process that operates on a population of "genotypes." The chapter describes the "metaperformance" process and the initial structure of the machines and also discusses the emergence of structured networks and computational abilities through strictly genetic adaptation. In his "genetic algorithms," Holland has introduced the notion of genetically adaptable programs. These programs consist of modules that can be combined in various ways, and that contain a variety of parameters with specified values in each individual program. By providing a mechanism for evolving new structure, the problems can be solved successfully by network-based computational systems.
