Parallel evolution of communicating classifier systems

We present an architecture that allows the division of a search space and the parallel solution of the resulting sub-problems. We use multiple genetic algorithms to evolve communicating classifier systems, where each classifier system represents a sub-system of the complete task. Any communication is uninterpreted and emergent to the system, indicating structure and interdependence between the sub-problems. A simulated trail following task, with three communicating classifier systems, is used to demonstrate the approach and we compare its performance to that of an equivalent single classifier system responsible for the whole problem.<>