Premium
Fusion of Intelligence Information: A Bayesian Approach
Author(s) -
PatéCornell Elisabeth
Publication year - 2002
Publication title -
risk analysis
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.972
H-Index - 130
eISSN - 1539-6924
pISSN - 0272-4332
DOI - 10.1111/0272-4332.00056
Subject(s) - probabilistic logic , computer science , bayesian probability , event (particle physics) , intelligence analysis , quality (philosophy) , artificial intelligence , information quality , terrorism , sensor fusion , signal (programming language) , information fusion , fuzzy logic , computer security , data mining , machine learning , information system , engineering , political science , programming language , philosophy , physics , epistemology , quantum mechanics , law , electrical engineering
The attack that occurred on September 11, 2001 was, in the end, the result of a failure to detect and prevent the terrorist operations that hit the United States. The U.S. government thus faces at this time the daunting tasks of first, drastically increasing its ability to obtain and interpret different types of signals of impending terrorist attacks with sufficient lead time and accuracy, and second, improving its ability to react effectively. One of the main challenges is the fusion of information, from different sources (U.S. or foreign), and of different types (electronic signals, human intelligence, etc.). Fusion thus involves two very distinct and separate issues: communications , i.e., ensuring that the different U.S. and foreign intelligence agencies communicate all relevant and accurate information in a timely fashion and, perhaps more difficult, merging the content of signals , some ``sharp'' and some ``fuzzy,'' some dependent and some independent into useful information. The focus of this article is on the latter issue, and on the use of the results. In this article, I present a classic probabilistic Bayesian model sometimes used in engineering risk analysis, which can be helpful in the fusion of information because it allows computation of the posterior probability of an event given its prior probability (before the signal is observed) and the quality of the signal characterized by the probabilities of false positive and false negative. Experience suggests that the nature of these errors has been sometimes misunderstood; therefore, I discuss the validity of several possible definitions.