On designing approximate inference algorithms for multiply sectioned Bayesian networks

An increasing number of applications require cooperative agents to reason about the state of an distributed uncertainty domain. However, inference process of such system could become overly slow for practical applications, and there has been significant interest in developing faster approximation techniques. In this paper, we focus on the existing MSBN models for cooperative reasoning in multi-agent environments. We show that, while the MSBNs provide a framework for exact inference, existing algorithms are usually not feasible in larger problem domains. Therefore, we investigate the issues related to the design of efficient inference algorithm for the MSBN model. We then propose a suite of algorithms for approximate multi-agent probabilistic reasoning in MSBNs. Our approach includes an MSBN subnet calibration process and distributed stochastic sampling on MSBN LJFs.