A generic architecture for hybrid intelligent systems

The integration of different learning and adaptation techniques in one architecture, to overcome individual limitations and achieve synergetic effects through hybridization or fusion of these techniques, has in recent years contributed to a large number of new intelligent system designs. Most of these approaches, however, follow an ad hoc design methodology, further justified by success in certain application domains. Due to the lack of a common framework it remains often difficult to compare the various systems conceptually and evaluate their performance comparatively. In this paper we first aim at classifying state-of-the-art intelligent systems, which have evolved over the past decade in the soft computing community. We identify four categories, based on the systems' overall architecture: (1) single component systems, (2) fusion-based systems, (3) hierarchical systems, and (4) hybrid systems. We then introduce a unifying paradigm, derived from concepts well known in the AI and agent community, as conceptual framework to better understand, modularize, compare and evaluate the individual approaches. We think it is crucial for the design of intelligent systems to focus on the integration and interaction of different learning techniques in one model rather then merging them to create ever new techniques. Two original instantiations of this framework are presented and discussed. Their performance is evaluated for prefetching of bulk data over wireless media.