KNOWLEDGE DRIVEN DATA COLLECTION USING LINE OF SIGHT EVIDENTIAL REASONING

Abstract

Data has been defined as low level individual observations and measurements. Related pieces of data can be organized to produce information. Information can be analyzed, understood, and explained to produce knowledge. The data leads us to knowledge. But can knowledge lead us to data? Is it possible to let the knowledge we desire drive the data we collect? This research presents a method to do just that. In this thesis I show how a simplified and abstracted Line of Sight Evidential Reasoning Analysis methodology can be used to drive evidence collection in three problem areas: verification and validation of modeling and simulation; sensor management; and information management. Line of Sight is defined as the ability to make a probabilistic inference about how collecting (or not collecting) a piece of evidence will influence our belief about a matter of interest. This is accomplished by building an evidence based Bayesian network which captures our belief about the matter of interest. A model has force if it is able to inform and positively influence decision makers. To a large extent, the force a model has on a decision maker is determined by the confidence the decision maker has in the model. Ideally that confidence relates to the quality of the verification and validation it has undergone. However, all verification and validation efforts are necessarily constrained by time and resources. This gives rise to the question ‘How can I best allocate limited verification and validation resources to increase the force a model has on a decision maker?’ This research presents a new quantitative method to rank verification activities based on how much force each will have on the decisions maker. A navy ship in a high threat area would ideally like to have perfect knowledge about its surrounds. This is called situational awareness. In practice perfect situational awareness is not possible; one reason for this is there are not enough sensor resources. This research shows how abstracted Line of Sight Evidential Reasoning analysis can be used to rank sensor tasking based on how much impact each tasking has on situational awareness. The internet of things brought an unprecedented number and variety of devices into a single network. This ubiquity presents opportunities and challenges for many types of systems including control systems. In the real world there may be bandwidth and reliability constraints which limit how much information can be transmitted. Given such limitations, this research demonstrates how abstracted Line of Sight Evidential Reasoning Analysis can be used to determine what information should be transmitted and from which source it should be obtained.