Introducing semi-supervised Support Vector Machines (S3VM)

Let's suppose we have a dataset made up of N labeled points and M (normally M >> N) unlabeled ones (a very common situation that arises when the labeling cost is very high). In a semi-supervised learning framework (for further details, please refer to Mastering Machine Learning Algorithms, Bonaccorso G., Packt Publishing, 2018), it's possible to assume that the information provided by the labeled samples is enough to understand the structure of the underlying data generating process. Clearly, this is not always true, in particular when the labeling has been done only on a portion of specific samples. However, in many cases, the assumption is realistic and, therefore, it's legitimate to ask whether it's possible to perform a full classification using only a poorly-labeled dataset. Of course, the...