Presenting the special issue on Rough-neuro computing : Preface

It goes without saying that a challenging quest for the construction of intelligent systems is realized through the development of hybrid information technologies and their vigorous and prudent exploitation. In a nutshell, what has emerged under the name of computational intelligence (CI) or soft computing is a well-orchestrated, highly synergistic consortium of technologies of neural networks, statistics, adaptive systems, granular computing and evolutionary methods. The knowledge is a multifaceted concept. So is the notion of CI as it attempts to take full advantage of the already mentioned technologies while compensating for some of their discrepancies or limitations. In a nutshell, we encounter a hybridization that looks into a matter of knowledge representation, uncertainty, and information granulation on one hand and an issue of learning, adaptation, and self-organization on the other. Granular computing including rough sets and fuzzy sets is about the "rst facet of hybrid intelligent systems. Neural networks are after the broad spectrum of learning. This issue of Neurocomputing is a testimony to this ongoing hybridization of the technologies of neural networks and granular computing, in particular rough sets and fuzzy sets. There are 10 papers addressing various methodological, algorithmic and application aspects of the hybrid approach. The paper by Andrzej Czyz5 ewski and Rafa" KroH likowski addresses an interesting problem of neuro-rough hybridization applied to digital processing of audio signals. The application of some selected techniques of CI to non-stationary noise reduction is also described. Pawan Lingras presents fuzzy-rough and rough-fuzzy serial combinations in neurocomputing. His paper introduces rough and neo-fuzzy neurons as well as the architectures of fuzzy-rough and rough-fuzzy subnets. Potential applications of the subnets are discussed together with illustrative examples. Evolutionary modular design of rough knowledge-based neural network is discussed in the article by Sushmita Mitra, Pabitra Mitra, and Sankar K. Pal. This article describes a way of integrating rough set theory with a fuzzy MLP using a modular evolutionary algorithm, for classi"cation and rule generation in soft computing paradigm. A rough set method is used for extracting dependency rules directly from a real-valued attribute table consisting of fuzzy membership values. An application of rough sets for enhancement of local subspace classi"ers is presented in the article by W"adys"aw Skarbek. Local subspace classi"ers are characterized by high performance. However, they lack a mechanism for recovery from errors. Con"dence levels associated with the decision function in this kind of classi"ers