|
Wim Hordijk |
ABSTRACT: A new class of models is developed which provides a tool for analyzing emergent computation in cellular automata (CAs). In particular, CAs that were evolved with a genetic algorithm to perform computational tasks requiring global information processing are analyzed. The models are used to make quantitative predictions of the computational performance, on a given task, of these evolved CAs. The models, and the resulting performance predictions, are based on the (emergent) dynamics of the evolved CAs and thus relate the dynamics of these CAs directly to their emergent computational abilities.
Furthermore, the class of models is used to determine quantitatively how and to what extent changes in the emergent dynamics of a CA give rise to changes in its performance. In particular, the differences between the dynamics of CAs that are related to each other in an evolutionary sense are analyzed this way. This, in turn, contributes to a better understanding of the evolution of emergent computation in CAs.
Finally, the class of models itself is investigated more thoroughly. For example, a correctness proof of the models is presented and an expression for scaling the models to larger system sizes is derived. The development, application, and investigation of this class of models thus forms a study of, and provides a better understanding of, the relation among dynamics, emergent computation, and evolution in cellular automata.