An Improved Architecture for Iterative Message Passing in MSBN

In this paper, we propose an improved architecture that supports exact MSBN belief updating using iterative message passing. Compared with recursive inference algorithms, iterative message passing is more robust in a multiagent environment where agents often have to face unreliable communication channels. Our method improves the overall time efficiency of existing iterative methods by avoiding repeated multiplications during message computation. Furthermore, in the proposed method, agents integrate incoming messages as batch at pre-determined intervals. We show that after enough message passings, an agentpsilas local belief will converge to the correct result.