Issues in training set selection and refinement for classification by a feedforward neural network

Training patterns are of unequal importance in image classification. For classification by a neural network, training patterns that lie close to the location of decision boundaries in feature space may aid the derivation of an accurate classification. The role of such border training patterns is investigated. A neural network trained with border patterns had a lower accuracy of learning but significantly higher accuracy of generalisation than one trained with patterns drawn from the class cores. Unfortunately, conventional training pattern selection and refinement procedures tend to favour core training patterns.