Single-Layer Networks

Abstract

Abstract In Chapter 1 we showed that the optimal decision rule for minimizing the probability of misclassification requires a new pattern to be assigned to the class having the largest posterior probability. We also showed how the posterior probabilities can be related to class-conditional densities through Bayes’ theorem, and in Chapter 2 we described several techniques for estimating these densities. An alternative approach, which circumvents the determination of probability densities, is based on the idea of a discriminant function, also introduced in Chapter 1. In a practical application of discriminant functions, specific parametrized functional forms are chosen, and the values of the parameters are then determined from a set of training data by means of a suitable learning algorithm.