Problem-Solving Approach to Data Fusion Alan Steinberg Space Dynamics Laboratory/Utah State University (asteinberg@sdl.usu.edu Abstract - This paper defines an approach for characterizing and solving data fusion problems in a system context. 1 We suggest a general ontology of problem-solving processes and characterize several types of data acquisition/fusion problems in terms of this ontology. By relating data acquisition/fusion problem solution to a more general theory of problem-solving, the methodology framework will assist the information system planner or analyst to • Find analogies in other domains to problems or sub- problems of interest • Learn from experience in solving analogous problems • Leverage techniques and metrics used in solving analogous problems. Keywords: Data fusion, data acquisition, problem- solving, sensor management, resource management, planning, information fusion, pattern recognition, learning systems, adaptive modeling. 1. Data Fusion and Problem-Solving In the most general sense, problem-solving involves planning and execution of actions. Often problem-solving involves elements of uncertainty. Such uncertainty can include: 1. Data Uncertainty - Lack of total confidence in available information (usually the result of imperfect information sources); 2. Model Uncertainty - Uncertainty in the expectations concerning the characteristics an behaviors of entities or situations; 2 3. Technique Uncertainty - Not knowing how to employ available resources to effect desired results; 1 The paper derives largely from effort performed as part of the Auspice Intelligence Research Committee for the U.S. Government. Some material appears in the final technical report for that project’s FY00 effort 2 Model uncertainty includes uncertainty regarding the characteristics of one’s own resources. Of particular interest is uncertainty as to the performance of one’s resources: the expected effects of candidate actions (probability of achieving an intended effect, or of unintended effects or of interactions among actions). 4. Goal Uncertainty - Lack of clear goals (i.e. incomplete, inconsistent, imprecise or fragile valuation of possible consequent states). Data fusion processes seek solutions to problems of a particular kind: estimation problems. Data fusion solves such problems by combining multiple data; e.g. by filtering commensurate data or by inferring characteristics that may not be directly observed. Data fusion is generally not performed in isolation, for the sake of data fusion itself. Rather, fusion is employed in the process of solving problems; its role being that of reducing at least some of the uncertainty factors in problem-solving. The first three of these uncertainty factors are attacked directly or indirectly in the process of estimating: 1. the state of the problem domain, 2. the fidelity of models of for predicting or interpreting characteristics of that domain, and 3. the effectiveness of techniques available to operate within that domain; including techniques for data acquisition and fusion. Furthermore, situational understanding can have a role in formulating and evaluating interim goals. It may influence even our highest-level goals (e.g. the value a person ascribes to selfish versus altruistic goals may vary with his perception of the situation). The problems that are expected to confront information system designers and information analysts are notable not only for their difficulty, but also for their diversity. These problems include those that are dominated by sensor phenomenology as well as those that focus on human and organizational activity and intent. We seek a general framework that will allow specific analytic methods to be developed and used effectively across this wide and expanding range of problems. Such a framework will facilitate the ability of planners and analysts to 1. Discover the significant characteristics of a new problem, using analytic and pattern-recognition techniques (Problem Pattern Discovery); 2. Recognize the similarities between the given problem and others that may have been solved already (Problem Pattern Recognition);