Incorporation of signal store into classifier systems: principle and experiments

To cope with dynamic learning environments, a signal store scheme is proposed to extend current classifier systems. Within our framework an intelligent agent leaves its own signals in the environment and later collects them to direct its learning process. Principles and components of the framework are outlined in the form of a “complete model”, followed by the introduction of the “current model” which is a preliminary implementation of the complete model. An experiment with the current model under a dynamic Woods1 environment is then introduced, together with discussions over its results. We conclude the paper by pointing out some possible improvements that can be made to the proposed framework