Detecting and Correcting Mistakes in Information Fusion; Conference paper

摘要

This article scrutinizes one inherent pitfall in automated information fusion - its fallibility as an artifact of dealing with the real world. Since it is highly desirable to avoid contamination of further processing by past mistakes, we investigate the nature of the recovery process in a prototype agent that performs the Level One Information Fusion task of entity tracking and re-identification, Smart-ASAS. Smart-ASAS attempts to solve this fusion task by treating it as one of Abductive Inference or Inference to the Best Explanation. We discover that the problem space bounding recovery from errors is exponential in nature, but investigate the possibility of handling this computationally complex problem with some proposed heuristics that would result in some satisfying solution.