This report presents research on various methods for heterogeneous information fusion — combining data that are qualitative, subjective, fuzzy, ambiguous, contradictory, and even deceptive, in order to form a realistic uncertainty-sensitive assessment of threat. The context is counterterrorism, for both military and civilian applications, but the ideas are more generally applicable in intelligence and law enforcement.
- How can analysts fuse complex information from different sources to form a realistic assessment of threat?
- What capabilities should a research prototype system for uncertainty-sensitive heterogeneous information fusion have?
- How can information fusion be accomplished despite numerous uncertainties about threat models, fusion mathematics (structural uncertainties), the data to be fused?
An element of thwarting terrorist attacks is observing suspicious individuals over time with such diverse means as scanners and other devices, travel records, behavioral observations, and intelligence sources. Such observations provide data that are often both complex and "soft" — i.e., qualitative, subjective, fuzzy, or ambiguous — and also contradictory or even deceptive. Analysts face the challenge of heterogeneous information fusion — that is, combining such data to form a realistic assessment of threat. This report presents research on various heterogeneous information fusion methods and describes a research prototype system for fusing uncertainty-sensitive heterogeneous information. The context is counterterrorism, for both military and civilian applications, but the ideas are also applicable in intelligence and law enforcement.
Table of Contents
Concepts, Methods, and a Research Platform
Creating Synthetic Data: Vignettes for Testing
Simple and Bayesian Fusion Methods
The Maximum Entropy/Minimum Penalty Fusion Method
Conclusions and Recommendations
A Factor-Tree Model for Propensity for Terrorism (PFT)
Extending Elementary Bayes Updating