Evolving cellular automata to perform computations. Final technical report

Abstract

The overall goals of the project are to determine the usefulness of genetic algorithms (GAs) in designing spatially extended parallel systems to perform computational tasks and to develop theoretical frameworks both for understanding the computation in the systems evolved by the GA and for understanding the evolutionary process which successful systems are designed. In the original proposal the authors scheduled the first year of the project to be devoted to experimental grounding. During the first year they developed the simulation and graphics software necessary for doing experiments and analysis on one dimensional cellular automata (CAs), and they performed extensive experiments and analysis concerning two computational tasks--density classification and synchronization. Details of these experiments and results, and a list of resulting publications, were given in the 1994--1995 report. The authors scheduled the second year to be devoted to theoretical development. (A third year, to be funded by the National Science Foundation, will be devoted to applications.) Accordingly, most of the effort during the second year was spent on theory, both of GAs and of the CAs that they evolve. A central notion is that of the computational strategy of a CA, which they formalize in terms of domains, particles, and particle interactions. This formalization builds on the computational mechanics framework developed by Crutchfield and Hanson for understanding intrinsic computation in spatially extended dynamical systems. They have made significant progress in the following areas: (1) statistical dynamics of GAs; (2) formalizing particle based computation in cellular automata; and (3) computation in two-dimensional CAs.