This study is concerned with the design and analysis of digital systems using fuzzy neurocomputing. Some basic definitions of fuzzy-logic neurons, fundamental neural architectures, learning strategies and interpretation of their results are presented. We promote a concept of embedding principle: an original Boolean problem is represented in the language of fuzzy sets, afterwards solved through learning, and, finally, the result of learning re-interpreted in terms of two-valued logic. We show the use of fuzzy neurons in a standard design of combinational systems including a minimization of incompletely specified Boolean functions (viz. those involving don't care conditions) and Boolean functions with many outputs. It is also claimed that this approach supports reverse engineering in the sense that once the architecture of the Boolean circuit has been learned, one can interpret the fuzzy logic network in order to gain an insight into the nature of rules governing the data.
This study is concerned with a fundamental issue of time series representation for modeling and prediction with Fuzzy Cognitive Maps. We introduce two distinct time series representation schemes for Fuzzy Cognitive Map design. First method is based on time series amplitude, amplitude change, and change of amplitude change (dynamics perspective). Second scheme is based on three consecutive historical observations: present value, past value and before past value (history perspective). Introduced procedures are experimentally verified and compared on several synthetic and real-world time series of various characteristics. The history-oriented time series representation turned out to be more advantageous. Quality of FCM-based time series models and one-step-ahead predictions were measured in terms of Mean Squared Error. We have shown that models designed with history-oriented time series representation generally require less FCM nodes to be of comparable quality as models built on dynamics-oriented time series representation. As a result, with the history-oriented time series representation scheme we are able to construct simpler and therefore better models.
We discuss various relationships of fuzzy sets and rough sets with knowledge discovery in databases (KDD). We describe the main advances of rough set and fuzzy set methods in solving KDD problems, point out research directions in fuzzy sets and rough sets stimulated by KDD, as well as characterize potential impact of these methodologies on KDD.
Fuzzy neurons may have outstanding learning abilities and are endowed with significant interpretation capabilities. In this study, we are concerned with the development of logic networks composed of fuzzy neurons. The main phase of the design includes the granulation of the output space (via triangular fuzzy sets) being realized with the use of fuzzy equalization. In the sequel these fuzzy sets are used to guide the construction of a family of fuzzy sets in the input space. Further processing of the resulting fuzzy sets deals with some additional aggregation of those that are not sufficiently distinct. This helps reduce the size of the logic network. We include comprehensive experimentation and offer a thorough interpretation of the networks. Experiments concerning real-world continuous data help evaluate the network's appealing properties: transparent interpretability and practical feasibility. © 2006 Wiley Periodicals, Inc. Int J Int Syst 21: 1249–1267, 2006.
The 1999 IEEE Canadian Conference on Electrical and computer Engineering (CCECE'99) was held from May 9 to 12, 1999, at the Shaw Conference Centre in Edmonton. The conference was a great success with over 380 papers presented and more than 400 peoples from 38 different countries presenting their recent research results. The area of Computational Intelligence was one of the vivid pursuits presented at the conference. Subsequently, we have been invited by the Editors-in-Chief of the Journal of Advanced Computational Intelligence to prepare a Special Issue of the Journal CCECE'99 conference. After a careful and strict peer review process, we have chosen six papers to be included in this special issue. They are selected from more than 20 papers submitted to this special issue, which are extended versions of the papers presented at the CCECE'99 conference in the areas of advanced computational intelligence. The papers fully reflect the breadth and diversity of conceptual and algorithmic facets of Computational Intelligence along with a spectrum of applications. We thank the authors and reviewers for doing an excellent job. We are grateful to Kaoru Hirota and Toshio Fukuda for making this selection of papers a part of the journal. We do hope the readers will enjoy this issue.
Deep learning is a timely research direction in machine learning, where breakthrough progress has been made in both academe and industries, bringing promising results in speech recognition, computer vision, industrial control and automation, etc. The motivation of deep learning is primarily to establish a model to simulate the neural connection structure of the human brain. While dealing with complex tasks, deep learning adopts a number of transformation stages to deliver the in-depth description and interpretation of the data. Deep learning achieves exceptional power and flexibility by learning to represent the task through a nested hierarchy of layers, with more abstract representations formed successively in terms of less abstract ones. One of the key issues of existing deep learning approaches is that the meaningful representations can be learned only when their hyperparameter settings are properly specified beforehand, and general parameters are learned during the training process. Until now, not much research has been dedicated to automatically set the hyperparameters, and accurately find the globally optimal general parameters. However, this problem can be formulated as optimization problems, including discrete optimization, constrained optimization, large-scale global optimization, and multiobjective optimization, by engaging mechanisms of evolutionary computation.
We are concerned with data mining in a distributed environment such as the Internet. As sources of data are distributed across the WWW cyberspace, this organization implies a need to develop computing agents exhibiting some form of collaboration. We propose a model of collaborative clustering realized over a collection of datasets in which a computing agent carries out an individual (local) clustering process. The essence of a global search for data structures carried out in this environment deals with a determination of crucial common relationships occurring across the network. Depending upon a way in which datasets are accessible and on a detailed mechanism of interaction, we introduce a concept of horizontal and vertical collaboration. These modes depend upon a way in which datasets are accessed. The clustering algorithms interact between themselves by exchanging information about "local" partition matrices. In this sense, the required communication links are established at the level of information granules (more specifically, fuzzy sets or fuzzy relations forming the partition matrices) rather than data that are directly available in the databases.
