In our approach, rather than produce control policies from a fixed or parametrized model, the KBMC system generates a decision model at run-time based on the problem description and information received thus far. Model construction consists of selection, instantiation, and assembly of causal and associational relationships from a broad knowledge base of general relationships among domain concepts. The system interleaves generation of model structure with analysis, so that the overall process can be directed toward issues relevant to the control options salient to the current decision context.
We are particularly interested in highly dynamic tasks, such as traffic assessment, where the focus of attention changes rapidly based on the information and control options available. In traffic assessment, for example, it is never feasible to evaluate a comprehensive traffic model at the highest level of fidelity and precision. Instead, our technique generates a coarser traffic model, and then uses the results of evaluating this model to determine which parts (e.g., which traffic sectors) should be modeled in more detail. By iterating this procedure, our KBMC system will converge on a traffic model that customizes the level of detail to the salient features and time demands of the particular problem at hand.
For a survey article on KBMC, see:
MP Wellman, JS Breese, and RP Goldman, From knowledge bases to decision models. Knowledge Engineering Review, 7(1): 35-53, 1992.A special section of IEEE Transactions on Systems, Man, and Cybernetics devoted to this topic will be published in November 1994.
Our work in this area is sponsored by the Air Force Office of Scientific Research, under Grant F49620-94-1-0027.