Efficient design and inference in distributed Bayesian networks: an overview

This paper discusses an approach to distributed Bayesian modeling and inference, which is relevant for an important class of contemporary real world situation assessment applications. By explicitly considering the locality of causal relations, the presented approach (i) supports coherent distributed inference based on large amounts of very heterogeneous information, (ii) supports a systematic validation of distributed models and (iii) can be robust with respect to the modeling deviations of parameters. The challenges of distributed situation assessment applications and their solutions are explained with the help of a real world example from the gas monitoring domain.