Towards the Development of Cognitive Maps in Classifier Systems

Classifier systems are well tested vehicles for implementing genetic algorithms in machine learning environments. This paper presents a novel system architecture that transforms a classifier system’s knowledge representation from message-based structures to self-organizing neural networks. These networks have been integrated with a classifier system to produce a Hybrid Learning System (HLS) that exhibits adaptive behaviour when driven by low level environmental feedback. Problems are represented within HLS as objects characterized by environmental features. Objects controlled by the system have preset goals set against a subset of their features and the system has to achieve these goals by developing a behavioural repertoire that efficiently explores and exploits the problem environment. Three types of knowledge structures evolve during this adaptive process: a cognitive map of useful regularities within the environment (encoded in a single self-organizing network); classifier behaviour calibrated against feature states and targets (encoded in a set of self-organizing feature maps); a population of complex behaviours (evolved from a gene pool supplied as part of the initial problem specification).