Meta-Patterns and Higher Order Meta-Patterns in Cellular Systems.

A new method for function approximation and classification in cellular systems is developed. The solution of a function approximation or a classification problem does not correspond to a connection matrix, but to a pattern in a cellular automaton. The pattern is obtained by a procedure of chaotic growth. It acts as a map between input- and ouput-patterns, and in this sense it is a meta-pattern. It is shown that problems with symmetry are typically solved by fractal patterns with aesthetic attractiveness. Also meta-meta-patterns are introduced, which are patterns that solve sets of problems instead of a single problem. Such patterns specify sets of sets of couples of input- and ouput-patterns. This schema is generalized straightforwardly for higher order meta-patterns. [ABSTRACT FROM AUTHOR]