A Taxonomy for Semi-supervised Learning Methods


Matthias Seeger
Max Planck Institute for Biological Cybernetics
Germany


We propose a simple taxonomy of probabilistic graphical models for the semi-supervised learning problem. Our aim is to point out some sources of confusion and suggest clarifications, rather than proposing a novel SSL method. We highlight differences and points of overlap between the proposed families and illustrate these using specific realizations in the litterature.