Special issue on Hybridization of Intelligent Systems

Hybridization of intelligent systems is a promising research field of computational intelligence focusing on synergistic combinations of multiple approaches to develop the next generation of intelligent systems. A fundamental stimulus to the investigations of Hybrid Intelligent Systems (HIS) is the awareness that combined approaches will be necessary to solve challenging problems in complex real-world situations. Neural computing, machine learning, fuzzy logic, evolutionary algorithms, agent-based methods, swarm optimization, quantum computing, are some intelligent systems – among others – that have been established and have shown their strengths and drawbacks. Giving the attribution of being “intelligent” to a system seems to be only admissible, if the system itself is sufficiently complex. However, all attempts to pose a structure on such complex systems show that the chosen structural components become intermingled, as the systems’ modeling itself tangents a practical application. If there is a layer model, the layers start to interact; if there are components, they cease to act independently from each other; and if there are rules, the rule conclusions will start to depend on context that is not captured by the rule antecedents. Compared to other systems’ features like dimension or scale, intelligence seems to be a vague system feature – it does not appear to be a property of a system that can be easily put into mathematical terms and equations, and whose presence can be easily verified by appropriately designed tests or experiments. Thus, intelligence is a cross-layer, cross-component and cross-rule specific matter, i.e. hybrid in its essence. In this special issue of the International Journal of Hybrid Intelligent Systems (IJHIS), we are presenting four excellent contributions, which demonstrate how intelligent systems emerge from hybridization. The focus is always on a complex system, be it the evolutionary design of image processing operators, the evolutionary reinforcement learning of neural networks, or the development of new general universally applicable learning strategies based on ensemble classification or clustering. The contributions are based on papers that were presented at the “Sixth International Conference on Hybrid Intelligent Systems”, which was held in conjunction with the “Fourth Conference on Neuro-Computing and Evolving Intelligence” (HIS-NCEI 2006) from December 13 to 15, 2006, at the Auckland University Technology Park in Auckland, New Zealand. About 70 papers devoted to the field of hybrid intelligent systems were presented at this conference. In the paper “Multi-objective Clustering Ensemble” by K. Faceli and M.C.P. de Souto, authors expand on recent works in multi-objective clustering and ensemble clustering. In the presented evolutionary approach, different individuals represent different clustering results, according to the use of several clustering methods with different parameter settings. New clustering results are produced from combination of existing ones in each generation, and multiple objectives are used for selecting among the various partitions. The proposed system is applied to synthetic data for verification, and to UCI benchmark data for proving efficiency. Similar to clustering ensembles, an increasing number of recent studies also considered the ensemble approach for classification. Particular attention was given to methods that combine a larger number of simple classifiers. Usually, these classifiers are specified in advance. In the paper “Genetic Rule Selection with a Multi-Classifier Coding Scheme for Ensemble Classifier Design” by Y. Nojima and H. Ishibuchi, authors are going a different way by actually generating the ensemble from a larger number of simple (interval-based) classifiers. The majority-voting results of the gener-