#### Foreword

Neural Networks, Fuzzy Logic and Evolutionary Computing are members of Soft Computing
class of techniques. The techniques are capable of identifying and handling inexact solutions
for complex tasks and can deal with the real life uncertainties within the computational
framework. Soft Computing has significantly matured over the years and we can find
significant applications of soft computing in industry and research environment. Neural
Networks is the most mature of other techniques in the Soft Computing. The networks have
also benefitted from integrated Fuzzy Logic based systems to model complex engineering
systems with the human expert knowledge and through robust system modelling.

Evolutionary Computing helps to optimise the design a Neural Network. Each members of the
Soft Computing has several algorithms and concepts that need better understanding for
application development.

This book on ‘artificial neural networks – principle and practice’ provides necessary
foundation to understand the basics of Neural Networks and how to develop real life
applications. From basic definitions to relevant theorems the book presents an algorithm
approach to describe the foundations.

The book also emphasises systematic approach to Intelligent System analysis and design. In
order to build a Neural Network or Artificial Neural Network application, one needs to apply
knowledge of probability based methods as well as fuzzy sets for more uncertain aspects of
the problem. The book then explains the motivation from our neural system to develop the
Neural Networks.

Description of other network-based approaches using nodes and edges also strengthens the
understanding about the Neural Networks. A major strength of the book is the fundamentals
of quantum logic for emerging Neural Network development. This is major area for future
development. A discussion on Neural Network hardware would have strengthened the book.

The second part of the book presents detailed discussion on learning algorithms, current and
emerging Neural Network structures and application development. The chapters also present
metrics to evaluate effectiveness of the network. Selection and integration of multiple Neural
Networks to solve a real life and complex problem is a major aspect of the book. As
mentioned before there are several algorithms and approaches to solve a problem, the
systematic approach presented in the book is of major interest. Application of the network to
solve a complex modelling task usually requires significant volume of data. A further
discussion on the modelling approaches with less data would be very helpful.

The emphasis on probability based neural network and its application is significant because of
its popularity. But the real strength of this part of the book is in describing the Quantum
Neural Networks and the Deep Belief Network (DBN).

Finally the book also outlines the research and development in Neural Networks. Future
Neural Networks are learning from specialised parts of our neural system and trying to scale
up to solve even more complex engineering applications.

**
***Rajkumar Roy*

Cranfield University, UK

#### Preface

An intelligent system is that which exhibit characteristics of learning, adaptation, and
problem-solving, among others. The group of intelligent system, conceived and designed by
human, is loosely termed Artificial Neural Network (ANN) System. Such ANN system is the
theme of the book. The book also describe nets (also called network or graphs), evolutionary
methods, clustering algorithm, and others nets, most of which are complementary to ANN
system.

The term “practice” in the title refers to design, analysis, performances assessment, and
testing. The design and analysis may be facilitated by the explanations, equations, diagrams,
and algorithms given. Performance assessments occur in any section that bear the name and
apply to any ANN system because they are standard independent methods and most ANN
system has an associated error feedback. Testing is exemplified by case studies and is given
toward the end of most chapters.

An interest in artificial neural sciences is a sufficient requirement to understand the content of
the book, though knowledge of signal processing, mathematics, and electrical/electronic
communication is an advantage. The book specifically takes a developmental perspective,
making it more beneficial for professionals. The book adopts a spiral method of description
whereby various topics are revisited several times; each visit introduces fresh material at
increasing level of sophistication. Each visit to a specific ANN type may also introduce new
ANN system(s) and/or new algorithm(s) as the case may be.

The book is divided into two parts (I and II). Part I contain five chapters. Chapter 1 introduce
the biological neurons and basic artificial neurons. From these, chapter 2 derive better
neurons and introduce statistical methods. Chapter 3 describe a framework of dynamic fuzzyneuron,
and explain the fundamental principle governing the design and analysis of ANN
system. To distinguish other algorithms (e.g. clustering algorithm) from learning algorithms,
chapter 4 describe fundamentals of genetic algorithm, clustering algorithms, and those other
algorithms complementary to ANN systems. Neural network is in chapter 3 introduced by
graph. Chapter 5 concludes part 1 by introducing quantum neural network, quantum maths
and logic. The chapter also describe Hodgkin-Huxley neuron, and memristance.

Similarly, part II consists of six chapters. In Chapter 6, artificial neuromorphic network, and
Widrow-Hoff learning are visited; so is fuzzy ANN system. While chapter 7 describes the
usual weighted, weightless ANN systems. It also introduces Bayesian ANNs, and discusses
general performance assessment methods. On the other hand, chapter 8 considers various
selection and combination strategy for ANN systems. Chapter 9 is dedicated to Bayesian networks. There are some promising ANN systems being considered in the research arena,
and also now in chapter 10, these ANN may revolutionize ANN throughput in future. In
chapter 11 implementation issues regarding Monte Carlo algorithm is visited, and also
implementation issues regarding neuromorphic networks is revisited.

The book attempts to impart considerable knowledge of know-how of ANN to the reader in
order to facilitate a novel development and research. Albeit also improve an ad-hoc ANN.
This may encourage and help a developer to meet any industrial increasing demand for novel
ANNs’ implementation and application.

**
***Pierre Lorrentz*

University of Kent,

United Kingdom

#### List of Contributors

##### Editor(s):

**Pierre Lorrentz **

University of Kent

United Kingdom