Improved Classification In Flat Networks
Date
2010-07-19
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Electrical Engineering
Abstract
It is shown that optimal flat networks can be found as solutions to least squares problems. An algorithm is presented to improve existing classifier training methods by changing the desired outputs. The algorithm is based on minimum probability of error. The algorithm's performance is compared with those of other algorithms including the Bayes Gaussian classifier. The Convergence of training and the effects of outliers are analyzed in all the algorithms presented here.