Classification systems based on neural networks

Classification is a problem that appears in many real life applications. In this paper we describe the general case of multi-class classification, where the task of the classification system is to map an input vector x to one of K>2 given classes. This problem is split in many two-class classification problems, each of them describing a part of the whole problem. These are solved by neural networks, producing an intermediate output in a reference space, which is then decoded to the solution of the original problem. The methods described here are then applied to the handwritten character recognition problem to produce the results described later in the article. It is suspected that they also may be applied successfully in the context of the CNN paradigm and be implemented on a CNN-Universal Machine.