Signature Table Systems and Learning

A characterization theorem is given for the classes of functions which are representable by signature table systems. The usefulness of the theorem is demonstrated in the analysis and synthesis of such systems. The limitations on the power of these systems come from the restrictions on the table alphabet sizes, and a technique is given for evaluating these limitations. A practical learning system is proposed and analyzed in terms of the theoretical model of this paper. Then an improved method is described and results are presented from a series of experiments.