A SCALABLE INFERENCING SYSTEM FOR CIVILIAN TERRORISM INTELLIGENCE

Abstract

This paper describes an approach to developing a scalable intelligence inferencing system for civilian terrorism intelligence. There is an obvious need for such a system in light of failures in the intelligence community leading to the September 11th attacks. It is intended as a supplement to human intelligence analysis; while intelligence analysts are good at what they do, it’s hard to see what information is important and integrate it. Information and inferences don’t always flow up the chain of command and don’t always get to where they are needed. Automated assistance can aid intelligence analysts, and managers helping to prevent other tragedies from occurring. The work explored an approach to automate (or provide assistance for) information fusion which effectively makes inferences over huge amounts of information and number of events using a scalable architecture. It is based on number of technical thought patterns: (1)evolutionary development of a system; (2)the use of layered inference graphs and tree based interpretations of them; (3) combining top-down with bottom-up inference; (4) and pattern matching with (un)certainty and importance calculations; (5)explanation based user interaction; and (6)using spatio-temporal localization, extrapolation/simulation and parallelism to raise inferencing performance to acceptable levels.