CellNet CoEv: Co-Evolving Robust Pattern Recognizers

An evolutionary model of classifier synthesis is presented. The CellNet system for generating binary pattern classifiers is used as a base for experimentation [5]. CellNet is extended to include a competitive co-evolutionary mechanism, where patterns (prey) evolve as well as pattern classifiers (hunters). This is facilitated by the addition of a set of topologically-invariant camouflage functions, through which patterns may disguise themselves. The addition of the co-evolutionary mechanism allows for a) the creation of a much larger and more varied pattern database (from the original), and also b) artificially increases the difficulty of the classification problem. Application to the CEDAR database of hand-written characters yields a) an increase in the accuracy and robustness of recognition, as well as b) the elimination of over-fitting, relative to the original CellNet software.