An Experimental Comparison of Hypothesis Management Approaches for Process Query Systems

A Process Query System (PQS) is a generic software system that can be used in tracking applications across a variety of domains. As in most other tracking systems, multiple hypotheses about which reports are assigned to which tracks must be maintained. Since the number of hypotheses that are possible can be exponential in the number of reports, some technique for managing a pool of the best candidate hypotheses must be used. In this paper, we compare a genetic algorithm approach and a hypothesis clustering approach with the basic top-H pruning policy. Metrics for comparison include performance accuracy and computational requirements. Simulations show positive results for both of these approaches and suggest that the clustering approach has the best overall performance. Other experiments indicate that the genetic algorithm technique can converge over time to the ground truth.