Reviews: Statistical Inference for Pairwise Graphical Models Using Score Matching

This paper studies the problem of estimating parameters in a pairwise graphical model and constructing confidence intervals for the parameters. As part of this an asymptotically normal estimator is constructed. The key progress made in this paper is that inference (i.e., confidence intervals) is done in a setting where computation of basic quantities (e.g. Specifically, an estimator based on Hyvarinen score is given for estimation of a single edge in the pairwise undirected graphical model. The new scoring rule uses the conditional density of two variables given the rest. A first step forms a preliminary Markov blanket for a pair of variables, and the estimator then re-optimizes over parameter value, which has a sort of decoupling effect.