Chapter 4. Semi-Supervised and Active Learning

In Chapter 2, Practical Approach to Real-World Supervised Learning and Chapter 3, Unsupervised Machine Learning Techniques, we discussed two major groups of machine learning techniques which apply to opposite situations when it comes to the availability of labeled data—one where all target values are known and the other where none are. In contrast, the techniques in this chapter address the situation when we must analyze and learn from data that is a mix of a small portion with labels and a large number of unlabeled instances.

In speech and image recognition, a vast quantity of data is available, and in various forms. However, the cost of labeling or classifying all that data is costly and therefore, in practice, the proportion of speech or images that are classified to those that are not classified is very small. Similarly, in web text or document classification, there are an enormous number of documents on the World Wide...