Combining Labeled and Unlabeled Data with Co-Training

<p>We consider the problem of using a large unlabeled sample to boost performance of a learning algorithm when only a small set of labeled examples is available.</p>