Lazy inference in multiply sectioned Bayesian networks using linked junction forests

Lazy propagation reduces the space complexity from HUGIN inference. Multiply Sectioned Bayesian Networks extend Bayesian Networks a cooperative multiagent paradigm. To combine the benefits of the two, a framework was proposed earlier to apply lazy propagation to inference in MSBNs. We propose an alternative framework with a simpler compiled structure. The issues of lazy communication and observation entering in a multiagent setting are considered. We prove that the inference is exact.