Efficient approximate inference in distributed Bayesian networks for MAS-based sensor interpretation

The multiply sectioned Bayesian network (MSBN) framework is the most studied approach for distributed Bayesian Network inference in an MAS setting. This paper describes a new framework that supports efficient approximate MASbased sensor interpretation, more autonomy and asynchrony among the agents, and more focused, situation-specific communication patterns. Its use can lead to significant improvements in agent utilization and time-to-solution.