Summary
The algorithms discussed in this chapter are generally more powerful than those analyzed in the previous one, but they have specific differences that must always be considered. CPLE and S3VM are inductive methods.
CPLE is an inductive, semi-supervised classification framework based on statistical learning concepts that can be adopted together with any supervised classifier. The main concept is to define a contrastive log-likelihood based on soft-labels that takes into account both labeled and unlabeled samples. The importance granted to the latter is conditioned to the maximization of the log-likelihood, and therefore the algorithm is less suited to tasks where fine control is needed.
Another inductive classification approach is provided by the S3VM algorithm, which is an extension of the classical SVM approach, based on two extra optimization constraints to address the unlabeled samples. This method is relatively powerful, but it's non-convex and, therefore, very...
