Integrated information processors with multi-functional components

In this paper, we propose integrated information processors with multi-functional and multi-level information processing components. These processors are introduced to simulate complex information processing supposed to be observed in living systems. Since it is difficult to model directly complex integrated information systems, we extract two types of information processing, that is, syntagmatic (sequential) and paradigmatic (parallel) processing. By combining these processing components, we approximate integrated, information systems of living systems. Information processors were applied to neural networks for information filtering and compressing. The performance of information processors was demonstrated by using a language acquisition problem. Experimental results confirmed that generalization performance can significantly be improved by controlling information, and that the inference mechanism is clearly interpreted.