Publications

In this article, we elaborate on a design methodology and the detailed realization of hierarchically structured granular models by engaging the fundamental principles and concepts of granular computing. The existing models are elevated to a more abstract (general) level by allocating a certain level of information granularity throughout the parameter space. In a concise way, the essence of the overall architecture of the proposed modeling mechanism could be generalized as follows: Numeric model (granular model of type-0)→granular model of type-1→granular model of type-2→…→granular model of higher type. The results of the granular models come in the form of type-0, type-1, or higher type information granules, which are decided by the overall level of hierarchy of the corresponding granular model. The specificity of granular outputs becomes a more comprehensive and sound quantification of the prediction accuracy and precision of the model and the quality of the specific prediction outputs. The proposed method facilitates effective communication with humans, who could get actively involved in the modeling process and determine the suitable level of abstraction depending upon the requirements of the problem. The determination of a suitable level of information granularity is realized with the guidance of the principle of justifiable granularity. A number of experimental studies concerning publicly available datasets are presented to illustrate the development methodology and show the effectiveness of the approach to form hierarchically structured solutions (reflecting different levels of abstraction) to the problem.

In this study, we introduce a new category of neurofuzzy networks – Fuzzy Polynomial Neural Networks and develop a comprehensive design methodology involving mechanisms of genetic optimization, and genetic algorithms, in particular. The augmented genetically optimized FPNN (referred to as gFPNN) is a structurally optimized architecture which comes with a higher level of flexibility in comparison to the one we have encountered in the conventional FPNN. The GA-based design procedure being applied to each layer of FPNN leads to the selection of preferred nodes (or FPNs) available within the FPNN. In the sequel, two general optimization mechanisms are explored. First, the structural optimization is realized via GAs whereas for the ensuing detailed parametric optimization is carried out in the setting of a standard least square method-based learning. The performance of the gFPNN is quantified through experimentation where we use a number of modeling benchmarks – synthetic and experimental data already experimented with in fuzzy or neurofuzzy modeling.

In this article, we propose a design and evaluation framework of granular neural networks realized in the presence of information granules. Neural networks realized in this manner are able to process both nonnumerical data, such as information granules as well as numerical data. Information granules are meaningful and semantically sound entities formed by organizing existing knowledge and available experimental data. The directional nature of mapping between the input and output data needs to be considered when building information granules. The development of neural networks advocated in this article is realized as a two-phase process. First, a collection of information granules is formed through granulation of numeric data in the input and output spaces. Second, neural networks are constructed on the basis of information granules rather than original (numeric) data. The proposed method leads to the construction of neural networks in a completely new way. In comparison with traditional (numeric) neural networks, the networks developed in the presence of granular data require shorter learning time. They also produce the results (outputs) that are information granules rather than numeric entities. The quality of granular outputs generated by our neural networks is evaluated in terms of the coverage and specificity criteria that are pertinent to the characterization of the information granules.

Fuzzy cognitive maps (FCMs) are a very convenient, simple, and powerful tool for simulation and analysis of dynamic systems. They were originally developed in 1980 by Kosko, and since then successfully applied to numerous domains, such as engineering, medicine, control, and political affairs. Their popularity stems from simplicity and transparency ofthe underlying model. At the same time FCMs are hindered by necessity ofinvolving domain experts to develop the model. Since human experts are subjective and can handle only relatively simple networks (maps), there is an urgent need to develop methods for automated generation of FCM models. This study proposes a novel learning method that is able to generate FCM models from input historical data, and without human intervention. The proposed method is based on genetic algorithms, and requires only a single state vector sequence as an input. The paper proposes and experimentally compares several different design alternatives of genetic optimization and thoroughly tests and discusses the best design. Extensive benchmarking tests, which involve 200 FCMs with varying size and density of connections, performed on both synthetic and real-life data quantifies the performance of the development method and emphasizes its suitability. © 2005 Elsevier B.V. All rights reserved.