While great progress has been made in the lowest levels of data fusion, practical advances in behavior modeling and prediction remain elusive. While there remains much room to improve data fusion algorithms, the most critical failing of existing approaches is their inability to support the required knowledge modeling and continuing refinement under constraints encountered in many real-world modeling problems, including the problem of modeling the effects of changing climate conditions on human behavior. These constraints include a lack of historic exemplars, knowledge distributed over many individuals, and the need to maintain complex models over long periods of time. The proposed Propheteer system will directly address this issue through a self-facilitating knowledge modeling approach that can flexibly exploit qualitative knowledge contributions made by individual SMEs, fuse these contributions with evidence from external data sources, and reason about multiple situation interpretations within a mathematically sound, quantitative framework of probabilistic inference. Finally, Propheteer will provide analysts with the insight required to recognize model shortcomings, and will exploit its ability to compose models from fragmentary knowledge to streamline ongoing model refinement. Our prior experience and existing software components will allow for rapid progress in Phase I prototyping and provide a solid foundation for a complete implementation in Phase II.
Keywords: Knowledge Elicitation, Probabilistic Inference, Information Fusion