Semi-supervised learning by disagreement

In real-world applications, assigning labels to examples usually requires human effort and therefore, labeled training examples are expensive; unlabeled training examples, however, are cheap and abundant. As a consequence, semi-supervised learning which attempts to exploit unlabeled data to help improve learning performance has become a very hot topic in machine learning and data mining. In this talk, I will introduce some of our research advances in disagreement-based semi-supervised learning, a paradigm covers a broad range of algorithms and has been successfully applied to many real tasks such as statistical parsing, noun phrase identification, image retrieval, etc.