During the past few years, semi-supervised learning has become a hot topic in machine learning and data mining, since manually labeling training examples is a tedious, error prone and time-consuming task in many practical applications. As one of the most predominant semi-supervised learning algorithms, co-training has drawn much attention and shown its superiority in many applications. So far, there have been a variety of variants of co-training algorithms aiming to settle practical problems. In order to launch an effective co-training process, these variants as a whole create their diversities in four different ways, i.e. two-view level, underlying classifiers level, datasets level and active learning level. This paper gives a review on co-training style algorithms just from this view and presents typical examples and analysis for each level respectively.
展开▼
