Tools for Facilitation of Influence Diagram Assembly in BMD Inference Collaborative Environment (BMD-ICE)

Construction and maintenance of large influence diagrams in an operational environment is a complex task. Modelers need tools to collaboratively build and evaluate these influence diagrams in the context of real world problems. Information Extraction andTransport (IET) is building the Ballistic Missile Defense Inference Collaboration Environment (BMD-ICE) to support specification of libraries of JPF frames, automatically generate situation specific influence diagrams and perform inference. Modelers needcapabilities to help them construct, interpret, debug, and extend these influence diagrams. To effectively validate a large influence diagram, a modeler needs to be able to focus on subnets and individual variables, to examine an enormous number ofprobabilistic relationships, to understand how situation-specific models evolve, and to understand the computational costs of his/her modeling decisions. We propose to design tools to aid the modeler in predicting, explaining, and controlling the behaviorof their models by supplying them with capabilities for contradiction checking, frame-based acyclicity checking, localization major influences on a given variable, subnet extraction, specification of test suites, stepthrough of model construction andinference, and complexity estimation. The result of this Phase I effort will be the analysis, design, and preliminary implementation of functionality to support modelers, in particular for localization of influence, frame-based acyclicity checking,frame-based contradiction checking, incremental step through, test suite execution, and influence diagram complexity estimation. The resulting technology will be applicable to markets that deal with large-scale diagnostics in uncertain environments (e.g.,aircraft maintenance, remote facilities management, factory maintenance, and real-time component monitoring), object recognition tasks in uncertain environments (e.g., data fusion, entity discrimination, and site monitoring), and event predictions incomplex and uncertain environments (e.g., identification of potential terrorist events and control of system inputs).