Information Fusion with Constrained Equivocation
MASSACHUSETTS INST OF TECH LEXINGTON LINCOLN LAB
Pagination or Media Count:
A review of information theory, statistical decision theory, and maximum entropy has resulted in a new technique for information fusion. Decision functions obtained from the traditional application of equivocation do not provide results that system designers will always be willing to accept this deficiency is resolved through the addition of constraints to the equivocation function. Statistical decision theory can then be shown to be a subset of the extended equivocation theory where the constraints fully define the solution to the problem. The extended theory indicates that a rich class of decision systems exists and that decision systems can be tailored to specific applications. The performance characteristics of the decision system are specified through the constraints as opposed to the ad hoc adjustments of the cost matrix that has often been used for traditional statistical decision theory. The extended theory indicates why some approaches to information fusion have been successful while others have not. The theory also shows that consistent constraints, or system objectives, are just as important as consistent a priori knowledge for the design of distributed information fusion systems. An information fusion formula has been obtained that is valid for a subset of equivocation constraints that meet certain requirements. The formula is expected to provide good results when fusing decisions from distributed equivocation-based decision systems. Reasonable results may also be attainable when fusing decisions from systems not based upon equivocation if those systems provide reasonable performance estimates.
- Information Science
- Statistics and Probability