Improved Generative Semisupervised Learning Based on Finely Grained Component-Conditional Class Labeling

We introduce new generative semisupervised mixtures with more fine-grained class label generation mechanisms than in previous works. Our models combine advantages of semisupervised mixtures, which achieve label extrapolation over a component, and nearest-neighbor (NN)/nearest-prototype (NP) classification, which achieve accurate classification in the vicinity of labeled samples. We propose several two-stage stochastic data generation mechanisms which involve first generating unlabeled data, then generating each labeled sample, not directly according to a component density, but rather based on unla-beled samples that were generated according to the component density. Variants of our method differ by whether both the samples and labels, or just the labels, are generated based on unlabeled samples. Our generation mechanisms entail more complicated (albeit still exact) E-step 1 evaluations than for standard mixtures. These form the basis for generalized EM algorithms for learning. Our models are advantageous when within-component class proportions are not constant over the feature space region " owned by " a component. Experiments on UC Irvine data sets demonstrate consistent gains in classification accuracy, compared with previous semisupervised mixtures. For clustering and density estimation, our methods outperform previous semisupervised approaches, but only outperform unsupervised mixture learning when there is both significant true component overlap and sufficient labeled examples to disambiguate components.