Improving Classification with Pairwise Constraints: A Margin-based Approach

In this paper, we address the semi-supervised learning problem when there is a small amount of labeled data augmented with pairwise constraints indicating whether a pair of examples belongs to a same