Evolutionary learning of temporal behaviour using discrete and fuzzy classifier systems

Abstract

We propose an architecture and representation, based on the learning classifier system, for the learning of temporal behaviour in intelligent agents operating in environments where reasoning about time, as well as space, plays an important part in the success of a learning agent. We draw our inspiration from two main biological sources: first, the Darwinian model of evolution, embraced by the genetic algorithm (GA) and second, the proposed existence of internal clocks in organisms for learning of period and interval timing. Biological evidence for internal clocks and their use in living organisms are briefly summarised. We describe two versions, discrete and fuzzy, of a novel learning classifier system which incorporates internal clocks for the express purpose of learning temporal behaviour. Several possible application areas of the proposed classifier system can be envisaged. These include intelligent control, using the classifier system either for direct control or as a temporal model; artificial life in environments with temporal as well as spatial characteristics; and temporal pattern recognition.