A Distributed Approach to Information Fu- sion Systems Based on Causal Probabilistic Models

In this paper we show that causal probabilistic models can facilitate design of robust and flexible fusion systems. Observed events resulting from stochastic causal processes can be modeled with the help of causal Bayesian networks, mathematically rigorous and compact probabilistic causal models. Bayesian networks explicitly represent conditional independence and this facilitates decentralized modeling and information fusion. Starting with the theory of BNs and factor graphs we derive design and organization rules for distributed multi-agent systems that implement exact belief propagation without centralized configuration and fusion control. We apply these design rules in distributed perception networks (DPN), an architecture that supports efficient and reliable fusion of large quantities of heterogeneous and uncertain information. DPNs consist of agents, processing nodes with limited fusion capabilities, which cooperate and organize themselves into complex distributed fusion systems for Bayesian network models. The DPN architecture supports design of robust and efficient monitoring and detection systems, which are helpful in contemporary decision making and controlling processes. Because of their modular design, DPNs can adapt to changing constellations of information sources (e.g. ad-hoc sensor networks consisting of mobile sensor suites or sensor motes) on the fly. Furthermore, their modular design facilitates distribution of fusion processes which can prevent processing and communication bottlenecks.