Manifold Regularization: A Geometric Framework for Learning from Labeled and Unlabeled Examples

We propose a family of learning algorithms based on a new form of regularization that allows us to exploit the geometry of the marginal distribution. We focus on a semi-supervised framework that in...