Ovates: Predictive Agents for Homeland Defense

ALPHATECH proposes to design and develop an approach to the identification and prediction of events from the semi-automated fusion and evaluation of information from multiple disparate data sources. The core of this proposal is the semi-automatic, ad hoc, and intelligent assembly of fragments of Bayesian networks to form adaptive belief networks that contain hypotheses about current and future events. The occurrence of events recorded by the data sources initiate the assembly of the belief networks, and additional data resident in the data sources are used by Ovates to evaluate the hypotheses and prune those with low likelihood of occurrence until only the most likely hypotheses about future events remain. The Ovates Bayesian blackboard, which supports the inferential predictive reasoning, is based on the classic blackboard architecture, extended for Bayesian inferential computations, agent-oriented knowledge sources, and a multi-agent architecture that provides access to the data sources. Problem and data source descriptions are based on semantics encoded in ontologies based on the DARPA Agent Markup Language (DAML) and the Ontology Web Language (OWL). Ovates is intended to predict the potential occurrence of terrorist threats within a community based upon local data sources such as emergency room, police, and first responder reports, traffic and police surveillance, and harbor activity reports.