Computational and learning synergies with a coevolving multilevel architecture

Abstract

Vertical information processing, including multi-level and multi-scale processing, plays an important role in biological systems since hierarchical structures might support learning and stability of the systems. The power of a vertical information processing system is more than the addition of the contribution of each constituting element together. Moreover, it means that the interactions of elements produce a synergistic effect that is greater than the sum of the individual elements. We have developed a coevolving computer model, motivated from physiological evidence, that integrates intra and interneuronal information processing, subjected to six levels of evolution. Information processing at the intraneuronal levels is to create a repertoire of pattern processing neurons. Information processing at the interneuronal levels is to group appropriate pattern processing neurons to constitute an effective pattern processing system. Evolutionary learning algorithms have been applied to each mode alternately. The system was tested with two sets of 1000 patterns. The experimental result shows that the system effectively employs synergy among different levels of information processing to obtain pattern differentiation capability. It also shows that synergies occur only when individual elements work with each other in a selectively cooperative manner.