The hot use-case of the PGMs approach has often been in the discriminative setting (see M^3 and latent SVM) - which is good because discriminative classifiers w

The hot use-case of the PGMs approach has often been in the discriminative setting (see M^3 and latent SVM) - which is good because discriminative classifiers work well with fewer data points (see Ng, Russel).