DECISION-THEORETIC APPROACHES TO SENSOR MANAGEMENT

KNOWLEDGE INDUSTRIES WILL EXPLORE PROBLEMS WITH REAL-TIME SENSOR MANAGEMENT, WITH A FOCUS ON FORMAL PROBABILISTIC AND DECISION-THEORETIC METHODS FOR REASONING UNDER UNCERTAINTY. FROM THE PERSPECTIVE OF DECISION ANALYSIS, WE VIEW SENSOR MANAGEMENT AS INCLUDING TWO KEY COMPONENTS: (1) ASSIGNMENT OF PROBABILITIES TO COMPETING EXPLANATIONS, GIVEN CURRENT SENSOR DATA; AND (2) DETERMINING WHICH SENSOR, OR SENSOR SUITE, SHOULD BE EMPLOYED NEXT TO REOLVE THE CURRENT UNCETAINITY. FOR THE PROBABILITY-ASSIGNMENT PROBLEM WE WILL INVESTIGATE THE FEASIBILITY OF EMPLOYING PROBABILISTIC CAUSAL NETWORKS CALL {\EM BELIEF NETWORKS} TO MODEL THE RELATIONSHIPS AMONG IMPORTANT VARIABLES FOR TACTICAL AIR FORCE AIRCRAFT OPERATIONS. SPECIFICALLY, WE WILL RESEARCH MODELS CONTAINING DIFFERENT MIXES OF DISCRETE AND CONTINUOUS VARIABLES, AND INVESTIGATE HOW DIFFERENT CONTEXTS AND TIME-DEPENDENCIES CAN BEST BE HANDLED. FOR THE SENSOR SELECTION PROBLEM, WE WILL EXPLORE THE USE OF ALTERNATE METHODS FOR REASONING ABOUT THE VALUE OF INFORMATION. IN PARTICULAR, WE WILL EXAMINE THE USE OF INFORMATION THEORY, DECISION THEORY, AND METHODS THAT COMBINE THE FORMAL APPROACHES WITH HEURISTIC CONTROL METHODOLOGIES. FOR THE PROBABILITY ASSIGNMENT AND SENSOR SELECTION PROBLEMS, WE WILL EVALUATE ALTERNATIVE APPROACHES WITH RESPECT TO COMPUTATIONAL EFFICIENCY, QUALITY OF DECISIONS, EASE OF MODEL CONSTRUCTION, AND MODEL MAINTENANCE. FINALLY, WE WILL EXAMINE REQUIREMENTS FOR A PHASE II EVALUATION AND IMPLEMENTATION.