Artificial Neural Systems: Principle and Practice

Pierre Lorrentz

• Publisher: Bentham Science Publishers
• Year: 2015
• Print ISBN: 9781681080918
• eBook ISBN: 9781681080901
• Topic: Computer Science
• Subtopic: Artificial Intelligence (AI) & Semantics
• Index: Computer Science

An intelligent system is one which exhibits characteristics including, but not limited to, learning, adaptation, and problem-solving. Artificial Neural Network (ANN) Systems are intelligent systems designed on the basis of statistical models of learning that mimic biological systems such as the human central nervous system. Such ANN systems represent the theme of this book. This book also describes concepts related to evolutionary methods, clustering algorithms, and others networks which are complementary to ANN system.
The book is divided into two parts. The first part explains basic concepts derived from the natural biological neuron and introduces purely scientific frameworks used to develop a viable ANN model. The second part expands over to the design, analysis, performance assessment, and testing of ANN models. Concepts such as Bayesian networks, multi-classifiers, and neuromorphic ANN systems are explained, among others.
Artificial Neural Systems: Principles and Practice takes a developmental perspective on the subject of ANN systems, making it a beneficial resource for students undertaking graduate courses and research projects, and working professionals (engineers, software developers) in the field of intelligent systems design.

Neural Networks

Pierre Lorrentz

Abstract

This chapter describe various types of ANN systems in relative detail. It is the aim of the chapter to give descriptions of advanced ANN system in such a detail as to facilitate easy implementation. The first few sections are dedicated to the recent weightless neural networks. This is followed by a weighted neural system section. Two advanced Bayesian network are introduced subsequently. The last section of the chapter explains the dynamics of ANN and how ANN nay be evaluated. The chapter has given a relatively extensive description of typical advanced neural networks from various categories of ANN systems.

Keywords: Adjustment, Back-propagation, Boltzmann distribution, Conditional probability, Division, Enhanced Probabilistic Convergent Network (EPCN), Generalized Likelihood Ration Test (GLRT), Helmholtz Machine, Kernel function, Kullback-Leibler divergence, Merging, Minimum Description Length, Mixture Density Network (MDN), Multi-classifier, Multi-expert System, Multi-Layered Perceptron (MLP), Probabilistic Convergent Network (PCN), Random Access Memory (RAM), Squared error, Wald test.

INTRODUCTION

In sections 1 and 2, some weightless neural networks are described in considerable detail. Weightless networks are presented here because they form a good alternative to weighted classical neural networks and also less prone to noise. A stereo-type weighted network, the Multi-Layered Perceptron (MLP), is described in the third section. The MLP is included because of its robustness and popularity; it represents a good example of weighted neural networks. Section four describes more advanced types of Bayesian classifier. They are suitably introduced here because the usual types of Bayesian classifiers have been described in chapter 6. The last section of chapter 7 presents the dynamics of an ANN system, and discussed the fusion mechanism of hierarchical network. This is followed by methods of independent evaluation of ANN systems.

The first and second sections of chapter 7 show ANN systems whose learning and recognition algorithms are derived from Boolean logic.. The PCN and EPCN may be employed in selection mechanism described in chapter 8. Seeking minimal sets of weights by MLP may be synonymous to seeking a set of (minimal) basis functions. Whatever the structural architecture of MLP that has been determined may be implemented by using the classical primitives (gates) of chapter 5. The MLP can be used as a component neural network of neuro-fuzzy system of chapter 6. The probability theories of chapters 2 may have provided sufficient background principle to the Bayesian networks of the fourth section of chapter 7. Any of the Bayesian network may participate in the selection mechanism of chapter 8. The last section of this chapter may be regarded as a continuation of performance evaluation methods that has been introduced in chapter 4. The performance evaluation mechanism of the last section is algorithmic and in considerable detail, whereas that of chapter 4 is introductory. Most ANN systems of other chapters may be evaluated for performance by using the methods of the last section of chapter 7.

Chapter 7 has described large number of standard ANN systems in considerable detail.

WEIGHTLESS NETWORKS

This section introduces a neural network whose functionality depends essentially on Boolean logic.

Probabilistic Convergent Network (PCN)

Many prediction problems and pattern recognition problems can be solved by performing Boolean logic on them. In situations whereby prediction or recognition problems can be interpreted in terms of Boolean logic, a type of random access memory (RAM) based network called Probabilistic Convergent Network (PCN) becomes suitable. An added advantage of PCN over existing RAM-based network is the inclusion of confidence measure.

To carry out a logic representing, it is anticipated that all inputs be condensed to threshold image. Due to the complexity of architecture and function of PCN, some terminologies are worth introducing w...

Table of contents

• Welcome
• Table of Contents
• Title Page
• BENTHAM SCIENCE PUBLISHERS LTD.
• FOREWORD
• PREFACE
• Principles
• Neurons
• Basic Neurons
• Basic Fuzzy Neuron and Fundamentals of ANN
• Fundamental Algorithms and Methods
• Quantum Logic and Classical Connectivity
• Practices
• Learning Methods
• Neural Networks
• Selection and Combination Strategy of ANN Systems
• Probability-based Neural Network Systems
• Emerging Networks
• Research and Developments in Neural Networks