EarthWeb   
HomeAccount InfoLoginSearchMy ITKnowledgeFAQSitemapContact Us
     

   
  All ITKnowledge
  Source Code

  Search Tips
  Advanced Search
   
  

  

[an error occurred while processing this directive]
Previous Table of Contents Next


Chapter 1
Preliminaries

1.1. Computational Intelligence: its inception and research agenda

Computational Intelligence (CI) (Bezdek, 1992) is a recently emerging area of fundamental and applied research exploiting a number of advanced information processing technologies. The main components of CI encompass neural networks, fuzzy set technology and evolutionary computation. In this triumvirate, each of them plays an important, well-defined, and unique role.


Figure 1.1  CI at a junction of neurocomputing, fuzzy sets and evolutionary computation

It is beneficial to look carefully at the main technological components of CI and identify their main objectives and research agendas.

Neural networks offer a powerful and distributed computing architecture equipped with significant learning abilities. They help represent highly nonlinear and multivariable relationships. Starting from pioneering research of Rosenblatt (1961) Minsky and Pappert (1969), neural networks have undergone a significant metamorphosis becoming today an important reservoir of various learning methods and learning architectures (Rumelhart and McLelland, 1986). They have already been successfully used in many system modeling, pattern recognition, robotics, and process control applications.

Fuzzy sets form a key methodology for representing and processing linguistic or, in general, non-numeric information. They support a diversity of mechanisms of knowledge representation focusing on a relevant selection of information granularity. Fuzzy sets exploit imprecision in an attempt to make system complexity manageable. There are two important principles supported by fuzzy sets. The first, formulated by the founder of fuzzy sets (Zadeh, 1973) is called the principle of incompatibility. In brief, it promotes fuzzy sets as a basic vehicle useful in overcoming an evident and acute disparity between precision and relevancy when modeling complex phenomena. The second principle originating within the realm of computer vision (Marr, 1982) alludes to an important idea of least commitment and graceful degradation that advises postponing any final decision until the point when enough evidence has been gathered. In this sense fuzzy sets allow quantification of uncertainty and take advantage of it rather than blindly discarding it.

Evolutionary computation embraces genetic algorithms (GA), evolutionary computation and evolutionary strategies which are biologically-inspired methodologies aimed at global optimization. The pioneering research of Holland (1975) and Fogel et al. (1966) gave rise to a new paradigm of population-driven computing.

All of these methodologies stem from essential cognitive aspects of fuzzy sets, underlying evolutionary mechanisms of genetic algorithms and biologically sound foundations of neural networks which provide essential foundations when dealing with engineering problems. With their increasing complexity, it becomes apparent that all of the technologies discussed above should be used concurrently rather than separately. Consider, for instance, the design of neural networks. Here, fuzzy sets deal with interfacing and preprocessing information in neural networks, especially if it comes in a nonumeric format. Evolutionary techniques are instrumental in determining not only the connections of the network but, more importantly, an entire topology of the network as well as its size. Another interesting example highlighting how CI contributes to an evident enhancement of the systems happens in the domain of rule-based computing. The rules are represented preliminarily throughout linguistic terms and afterwards augmented with the use of evolutionary techniques and neurocomputing, Fig. 1.2.


Figure 1.2  Rule-based computing: an evolution

The important observation is that the CI methodology retains its generality while being flexible enough to address the needs and specificity of particular applications. We can illustrate this synergistic effect in Fig.1.3. It reveals some relationships between fuzzy sets, neural networks and evolutionary computation and various application areas.


Figure 1.3  Methodological hub of Computational Intelligence

More importantly, CI provides us with another look at complex problems that otherwise would not have been able to approach and address when exploiting individual methodologies.

The key objective of the book is to expose the reader to the emerging area of CI and discuss its methodology along with the ensuing algorithms. Evidently, the field of CI is in stati nascendi; undoubtedly new ideas will be springing up and we will witness an emergence of new subareas.

1.2. Organization and readership

The material is organized into 8 chapters. The concise roadmap as outlined in Fig. 1.4 highlights the three basic methodological streams. We start off with a brief introduction to neural networks, fuzzy sets and evolutionary computing. These basic topics are covered in Chapter 2, 3, and 5. Essentially, these three chapters build up a solid prerequisite for any other tour despite its final destination. Chapter 4 reinforces the ideas of fuzzy sets in the applied setting by studying fuzzy set oriented computation including commonly used rule-based architectures. Chapter 7 is devoted to neuro-fuzzy architectures. In addition to a number of interesting topologies, the material includes in-depth coverage of the main methodological aspects of synergy between the technology of fuzzy sets and neurocomputing. In particular, we study the key issues of knowledge representation and learning - the two important facets to be considered inside out in the design of any CI construct. In Chapter 8, we discuss all the technologies put together and identify several factors that make a notion of evolutionary neuro - fuzzy systems (CI architectures) a reality. Each chapter in this part includes not only the detailed algorithms but elaborates on the methodological issues of hybridization. The book is kept self-contained to a high degree. We do not assume any previous exposure to any of these three technologies. In our opinion, it is primordial to start from scratch and recast the main methodologies, concepts and algorithms of genetic, fuzzy and neural computing in the setting of CI. As we have already emphasized, these technologies have their own research agendas and these need to be thoroughly reconsidered in the light of their emerging synergy.


Previous Table of Contents Next

Copyright © CRC Press LLC

HomeAccount InfoSubscribeLoginSearchMy ITKnowledgeFAQSitemapContact Us
Products |  Contact Us |  About Us |  Privacy  |  Ad Info  |  Home

Use of this site is subject to certain Terms & Conditions, Copyright © 1996-2000 EarthWeb Inc. All rights reserved. Reproduction in whole or in part in any form or medium without express written permission of EarthWeb is prohibited. Read EarthWeb's privacy statement.