Linear classifiers in perceptron design

Publisher Website
Google Scholar
Abstract
It is shown adaptive training of the nonlinear single layer perceptron can lead to seven different statistical classifiers: (1) Euclidean distance classifier; (2) standard Fisher linear discriminant function; (3) Fisher linear discriminant function, with pseudoinverse of the covariance matrix; (4) regularised discriminant analysis; (5) generalised Fisher discriminant function; (6) minimum empirical error classifier; and (7) maximum margin classifier and to intermediate ones. Which particular type of the classifier will be obtained depends on: 1) initialisation interval and its relation to the training data; 2) an initial value of the learning step; and 3) its change during the iteration process, the stopping criteria.

This publication has 15 references indexed in Scilit: