Automatic modularization by speciation

Real-world problems are often too difficult to solve by a single monolithic system. There are many examples of natural and artificial systems which show that a modular approach can reduce the total complexity of the system while solving a difficult problem satisfactorily. The success of modular artificial neural networks in speech and image processing is a typical example. However, designing a modular system is a difficult task. It relies heavily on human experts and prior knowledge about the problem. There is no systematic and automatic way to form a modular system for a problem. This paper proposes a novel evolutionary learning approach to designing a modular system automatically, without human intervention. Our starting point is speciation, using a technique based on fitness sharing. While speciation in genetic algorithms is not new, no effort has been made towards using a speciated population as a complete modular system. We harness the specialized expertise in the species of an entire population, rather than a single individual, by introducing a gating algorithm. We demonstrate our approach to automatic modularization by improving co-evolutionary game learning, learning to play the iterated prisoner's dilemma. We review some problems of earlier co-evolutionary learning methods, and explain their poor generalization ability and sudden mass extinctions. The generalization ability of our approach is significantly better than past efforts. Using the specialized expertise of the entire speciated population though a gating algorithm, instead of the best individual, is the main contributor to this improvement.