Representational and Inferential Requirements for Diagrammatic Reasoning in the Entity Re-Identification Task

摘要

Diagrammatic reasoning is ubiquitous in Army reasoning: situation understanding and planning in the Army both involve representing aspects of the situation and plans in the form of diagrams. We have been developing a general architecture to support diagrammatic reasoning for Army applications, and in an earlier report 1 we discussed an application in simple maneuver recognition. Our research strategy has been to investigate a variety of applications, each bringing additional requirements for perception and diagrammatic object creating capabilities that can assist the army. One such area is the army's All Source Analysis System (ASAS) that is designed to automate the processing and analysis of intelligence data from all possible sources. One of its purposes is to keep track of various enemy assets, based on a variety of sensors and reports, including direct sightings of entities. In order to maintain a coherent view, the system has to decide whether a new sighting refers to a new entity, a previously known entity that has since moved (re-identification) or is erroneous. This is the entity re-identification task. We have a built a system that uses an abductive reasoning process together with a diagrammatic reasoning system to solve this problem. In this paper, we look at some of the issues that the entity re-identification task poses for diagrammatic reasoning.