| Multiply Sectioned Bayesian Networks (MSBNs) is an extension of Bayesian networks and provides a framework for probablistic reasoning in multi-agent interpretation systems. A multi-agent MSBN may be constructed by independent vendors. Thus, to ensure the correctness of agent interface, automatic verification of d-sepset condition is desirable when an MSBN is composed. Furthermore, inference in an MSBN can be performed effectively using a compiled representation called a linked junction forest (LJF). Cooperative triangulation is an essential step in the compilation procedure. The problem of cooperative d-sepset verification and triangulation becomes difficult with the vendors' privacy being a necessary consideration.; This thesis presents and implements the algorithms for cooperative d-sepset verification and triangulation without violating the vendors' privacy. The generic framework for communication in multi-agent MSBNs is improved. The experimental results of the distributed MSBN verifier and compiler implemented show the effectiveness of the algorithms. |
