Gaussian conditional random fields for classification

Gaussian conditional random fields (GCRF) are a well-known used structured model for continuous outputs that uses multiple unstructured predictors to form its features and at the same time exploits dependence structure among outputs, which is provided by similarity measure. In this paper, binary classification (GCRFBC) proposed. The applicable problems with undirected graphs, intractable standard CRFs. representation of GCRFBC extended latent variables yield some appealing properties. Thanks GCRF structure, becomes tractable, efficient open improvements previously applied regression models. addition, allows reduction noise, might appear if structures were defined directly between discrete outputs. Additionally, two different forms algorithm presented: GCRFBCb (GCRGBC - Bayesian) GCRFBCnb (GCRFBC non Bayesian). method local variational approximation sigmoid function solving empirical Bayes in Bayesian variant, whereas MAP value basis learning inference variant. solved Newton-Cotes formulas one-dimensional integration. Both models evaluated on synthetic data real-world data. It was shown both achieve better prediction performance than predictors. Furthermore, computational memory complexity evaluated. Advantages disadvantages proposed discussed detail.