Generating Dependence Structure of Multiply Sectioned Bayesian Networks

Multiply sectioned Bayesian networks (MSBNs) provide a general and exact framework for multi-agent distributed interpretation. To investigate algorithms for inference and other operations , experimental MSBNs are necessary. However, it is very time consuming and tedious to construct MSBNs manually. In this work, we investigate pseduo-random generation of MSBNs. Our focus is on the generation of MSBN structures. Pseduo-random generation of MSBN structures can be performed by a generate-and-test approach. We expect such approach to have a very low probability of generating legal MSBN structures that satisfy all the technical constraints, and hence will be inefficient. We propose a set of algorithms that always generates legal MSBN dependeuce structures.