Bayes' Nets for the Dynamic Database

recent advances in sensor technology have enabled collection of vast quantities of battlefield information. Warfighters will soon gain global access to this huge volume of information. Rather than construct separate redundant, and inconsistent situation estimates, a Dynamic Database can reduce the data to a comprehensive, self-consistent representation of the environment. This proposal explores the use of Bayes' Nets as a central building block of a Dynamic Database. The key to a Dynamic Database is the maintenance of a statistically consistent set of hypotheses about the situation. Bayes' Nets provide a rigorous methodology for enforcing statistical consistency among a large set of parameter estimates and relational hypotheses. Technical challenges include the ability to efficiently update estimates in the database, incorporation of efficient search mechanisms, embedding of computational services, and presentation of products to users. A small end-to-end prototype will provide a complete context for the Bayes' Net investigation, and will identify system-level technical challenges relative to the representation. Phase II will develop prototype algorithms for the functions which interact with the situation model, and extend the prototype Dynamic Database to a detailed 10 km x 10 km situation model.