A practical neural network for handwritten character recognition built by dynamics-based active learning and self-organization of feedback mechanism

A feedback mechanism which enhances similarity between recognition by a human and that by an artificial neural network we have proposed is introduced. A human may turn his attention to a particular portion of the object in order to classify it successfully. It is regarded as a feedback from the first rough classification to the final one. The proposed feedback mechanism is designed to simulate such attention. The recognition rates by a neural network with the feedback mechanism exceed those by a former neural network.