Introduction to Semi-Supervised Learning

Semi-supervised learning is a machine learning branch that tries to solve problems with both labeled and unlabeled data with an approach that employs concepts belonging to clustering and classification methods. The high availability of unlabeled samples, in contrast with the difficulty of labeling huge datasets correctly, drove many researchers to investigate the best approaches that allow extending the knowledge provided by the labeled samples to a larger unlabeled population without loss of accuracy. In this chapter, we're going to introduce this branch and, in particular, we will discuss:

• The semi-supervised scenario
• The assumptions needed to efficiently operate in such a scenario
• The different approaches to semi-supervised learning
• Generative Gaussian mixtures algorithm
• Contrastive pessimistic likelihood estimation approach
• Semi-supervised Support Vector Machines (S³VM)
• Transductive Support Vector Machines (TSVM)