Quick Overview

Abstract

The main objective of this chapter is to define the terminology to be used and the primary issues to be considered in depth in the chapters that follow. In the first section, I present three of the simplest and most popular classification algorithms, and in the second section, I describe single and multilayer perceptrons and the training rules that are used to obtain classification algorithms. I give special attention to the most popular perceptron-training rule, back-propagation (BP). In the third section, I show that the three statistical-based classifiers described in Section 1.1 can be obtained via single-layer perceptron (SLP) training. The fourth section deals with performance, complexity and training-set size relationships, while the fifth section explains how to utilise these relationships while determining the optimal complexity of the decision-making algorithm. In the final section, I explain the overtraining effect that can arise while training artificial neural networks. Some bibliographical and historical remarks are included at the end of the chapter.