Rigorous Framework for Distributed Processing with Dynamic Stochastic Information

The vast and ever-increasing amount of networked information available creates a growing need for information processing systems that distribute decision making rather than centralize data. Under this program, Prism Informatix is developing a novel processing framework based on a rigorous theory of distributed cooperative (team) decision making under uncertainty called Distributed Hypothesis Testing Networks (DHTN). The approach is analytically tractable, extremely bandwidth efficient, and gives rise to solutions which are both intuitive in structure and optimal in a well-defined sense, all essential attributes in providing a path to scalable and analyzable distributed processing architectures. In Phase I of our program, a core DHTN reasoning engine was instantiated in software and evaluated extensively via simulation to establish its favorable performance characteristics for decision making accuracy, computational complexity, and robustness to practical processing constraints resulting from unreliable network performance. In Phase II, we propose to develop the capability into a full distributed system prototype in which distributed services cooperate to optimize global network-level decision metrics. We will demonstrate the operation of the prototype in two high-value net-centric processing applications: a multisource fusion application for the Navy and a distributed data mining application targeted for the private sector.