Incremental Learning of Nde Signals with Confidence Estimation

An incremental learning algorithm, Learn—, is introduced, for learning additional information from new data, even when new data include examples of previously unseen classes. Learn++ takes advantage of synergistic generalization performance of an ensemble of simple classifiers, each trained with a strategically chosen subset of the training database. As new data become available, new classifiers are generated, which are then combined through weighted majority voting. The weights are determined based on the estimated likelihood of each classifier to correctly classify an instance of unknown class. The voting procedure also allows Learn— to estimate the confidence level in its own decision.