Decision Making under Probability Intervals

During Phase I we have developed successful algorithms to handle probability interval propagation and second-order uncertainty (SOU) in Bayesian Networks (BNs). Building on this success, we have three goals for Phase II. First, we will refine and extend the algorithms we have developed in Phase I. This will be achieved through mathematical research that will lead to more efficient, precise and well-tuned algorithms. Second, we will integrate these algorithms in an Application Programming Interface (API). Third, we will build a development framework and a Graphical User Interface (GUI) to allow a user to intuitively develop SOU models. Our API will provide full support to Courses of Action generation with Probabilistic Relational Models, which allow the encoding of complex models that are impossible to handle with standard BNs. The GUI will feature novel techniques for eliciting the probabilistic parameters of a model, thus increasing the users' effectiveness and efficiency in encoding the knowledge. Finally, we propose to validate our development system and SOU algorithms with an MDA problem of interest selected from on-going Ballistic Missile Defense System programs. In this regard, we will specifically collaborate with the prime contractors of MDA's Project Hercules and Battlespace Environment and Signatures Toolkit (BEST) project.