A NOTE ON A LARGE MARGIN PERCEPTRON ALGORITHM

Abstract

The importance of classification algorithms in the context of risk assessment is briefly explained. As an alternative to the popular support vector machines fault tolerant perceptron learning is suggested. In order to achieve better generalization properties the additional use of an iterative large margin perceptron algorithm is investigated. In particular it is shown that care has to be taken when initializing the algorithm. Some preliminary experimental results are briefly discussed.