Learning with Graphical Models

Graphical models provide a powerful framework for probabilistic modelling and reasoning. Although theory behind learning and inference is well understood, most practical applications require approximation to known algorithms. We review learning of thin junction trees–a class of graphical models that permits efficient inference. We discuss particular cases in clique graphs where exact inference is possible in polynomial time and some special cases where good approximation guarantees can be given. We also point out the drawbacks of learning with approximate inference. Finally, a practical application of probabilistic generative model for learning visual attributes from images is discussed.