Learning Convex Inference of Marginals

Justin Domke

Graphical models trained using maximum likelihood are a common tool for probabilistic inference of marginal distributions. However, this approach suffers difficulties when either t…