Sampling from Gaussian graphical models using subgraph perturbations

Abstract

The problem of efficiently drawing samples from a Gaussian graphical model or Gaussian Markov random field is studied. We introduce the subgraph perturbation sampling algorithm, which makes use of any pre-existing tractable inference algorithm for a subgraph by perturbing this algorithm so as to yield asymptotically exact samples for the intended distribution. The subgraph can have any structure for which efficient inference algorithms exist: for example, tree-structured, low tree-width, or having a small feedback vertex set. The experimental results demonstrate that this subgraph perturbation algorithm efficiently yields accurate samples for many graph topologies.