Anytime hybrid propagation in Bayesian networks

Inference in Bayesian networks often gets prohibitively complex and one must resort to approximate inference methods.As the complexity of inference in Bayesian networks is most often determined by a few large cliques of the junction tree (used as the computational structure for the inference), a hybrid propagation algorithm, mixing exact and approximate message passing in the junction tree, seems like an obvious idea, which was explored by Dawid et al. (1995), and subsequently implemented in Hugin (Kjærulff, 1995). The current project aims at improving this methodology to allow anytime behaviour, making the inference process able to deliver prompt and continuously improving results as more time is allowed for the inference process.