We introduce Learn++, an algorithm for incremental training of neural network (NN) pattern classifiers. The proposed algorithm enables supervised NN paradigms, such as the multilayer perceptron (MLP), to accommodate new data, including examples that correspond to previously unseen classes. Furthermore, the algorithm does not require access to previously used data during subsequent incremental learning sessions, yet at the same time, it does not forget previously acquired knowledge. Learn++ utilizes ensemble of classifiers by generating multiple hypotheses using training data sampled according to carefully tailored distributions. The outputs of the resulting classifiers are combined using a weighted majority voting procedure. We present simulation results on several benchmark datasets as well as a real-world classification task. Initial results indicate that the proposed algorithm works rather well in practice. A theoretical upper bound on the error of the classifiers constructed by Learn++ is also provided.

A statistical approach for the recognition of Arabic characters is introduced. As a first step, the character is segmented into primary and secondary parts (dots and zigzags). The secondary parts of the character are then isolated and identified separately, thereby reducing the number of classes from 28 to 18. The moments of the horizontal and vertical projections of the remaining primary characters are then calculated and normalized with respect to the zero-order moment. Simple measures of the shape are obtained from the normalized moments. A 9-D feature vector is obtained for each character. Classification is accomplished using quadratic discriminant functions. The approach was evaluated using isolated, handwritten, and printed characters from a database established for this purpose. The results indicate that the technique offers better classification rates in comparison with existing methods.>