A Revised Approach to Combining Linguistic and Probabilistic Information in Correlation
Abstract
With the advent of the field of Artificial Intelligence (AI), new insights have been gained into information handling in general, and target data association, in particular. It is now clear that the traditional use of only stochastic geolocation information (i.e., vector positions, velocities, etc.) is not really adequate to describe the full data association problem. Rather, the situation can be markedly improved by using previously disgarded information in the form of linguistic descriptions or narratives, such as 'probably very long', 'appears to be flying a dark-blue flag', is rather oblong in appearance, etc. However, at present, despite the inroads made with AI techniques, there is no adequate comprehensive theory covering the complete modeling of such information in conjunction with the classical measurement information of geolocation. A previous attempt at addressing the above issue was formalized through the development of the PACT (Possibilistic Approach to Correlation and Tracking) algorithm developed at NOSC. The novelty of the algorithm was its explicit use of both linguistic-based evidence and probabilistic information through a common structure of a possibilistic/fuzzy-set model. The algorithm is essentially a generalization of a conditional form of the well-known total probability expansion theorem, an alternative to Bayes' theorem, when priors are not readily determinated; and a number of auxiliary attributes must be utilized to connect the parameter of interest-here, data association or correlation-with the observed data. Correlation, Fuzzy sets, Data association, Possibility functions, Knowledge-based systems, Multi-values logic.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 01, 1992
- Accession Number
- ADA269612
Entities
People
- I. R. Goodman