Prototype selection rules for neural network training

摘要

Rules to select a set of training prototypes from a collection of training prototypes are developed so that a neural network classifier converges to a solution when pattern classes overlap in feature space. The rules are especially useful for selecting training prototypes in order to improve the network robustness and operational flexibility by retraining the network with noisy prototypes. The formulation of the selection rules are based on a distortion measure and the network response to the training prototype collection. The application and effectiveness of the selection rules are demonstrated on a synthetic pattern classification in a Gaussian noise problem and a practical automatic target recognition problem.