Introduction: Hybrid intelligent adaptive systems

This issue of International Journal of Intelligent Systems includes extended versions of selected papers from the 4th International Conference on Soft Computing, held in Iizuka, Japan, September 30]October 5, 1996. The topic of the special issue is ‘‘Hybrid Intelligent Adaptive Systems.’’ Research on hybrid systems is one of the key issues of developing intelligent systems and it can apply a wide range of tools, including artificial neural networks, fuzzy logic, knowledge-based systems, genetic algorithms, evolutionary computation, and chaos models. The papers in this issue have been carefully reviewed and modified to give the readers a comprehensive overview of theoretical aspects, design, and implementation issues of hybrid intelligent adaptive systems. In the first paper by Kasabov and Kozma, a general framework of developing hybrid, intelligent, and adaptive systems is given. This work develops multimodular, fuzzy neural network systems and applies it to phoneme-based speech recognition. The second paper by Miyata, Furuhashi, and Uchikawa proposes fuzzy abductive inference with degrees of manifestations. This method infers irredundant combinations of candidates with degrees of belief for the manifestations. It is also demonstrated that the results of the inference method are applicable to medical diagnosis and system fault detection. Cho adopts ideas of artificial life in his work to develop evolutionary neural networks. The introduced modular neural network can evole its structure autonomously using a structural genetic code. The effectiveness of the method is demonstrated on the example of handwritten digit recognition. Feuring and Lippe study theoretical aspects of fuzzy neural networks. They propose a training algorithm for fuzzy neural networks that satisfies a certain goodness criterion. The second part of the special issue contains articles related to time series analysis and systems control. Yamazaki, Kang, and Ochiai introduce a hierarchical neural network system for adaptive, intelligent control, based on the analogy with the human thinking process. The optimum parameter space of the neural network system is found by a self-controllable algorithm, which can lead to either equilibrium or to nonequilibrium, chaotic behavior. The results of this study are applied, e.g., to laser beam analysis, semiconductor design, and design of magnetic devices. The work by Kozma, Kasabov, Kim, and Cohen presents a chaotic neuro-fuzzy method for time series analysis and process control. The