In this article we present a new class of support vector machines for binary classification task. Our support vector machines are constructed using only two support vectors and have very low Vapnik-Chervonenkis dimension, so they generalize well. Geometrically, our approach is based on searching...