Sparse inference for binary graphical models

Graphical models have recently regained interest in the statistical literature for describing associations among many random variables. The models discussed here use an undirected graph to encode conditional independence relationships among binary random variables. This study evaluates and proposes statistical methods for binary graphical models for problems in high dimensions and possibly low sample size. Two approaches are considered. First, the binary graphical model is formulated as a generalized linear model and regularized estimation is applied. Second, a latent multivariate Gaussian model is used.