A URREF interpretation of Bayesian network information fusion

In order for the uncertainty representation and reasoning evaluation framework (URREF) ontology for the evaluation of information fusion systems to have maximum value, it must be generally applicable irrespective of the application, uncertainty representation, reasoning scheme or data format. Since the URREF ontology is still an evolving framework, it is the focus of ongoing refinement through the efforts of the Evaluation of Techniques for Uncertainty Representation Working Group (ETURWG). Recent efforts by the authors to apply the URREF definitions to the evaluation of Bayesian network (BN) fusion systems have identified a need to translate and map the terminology of the URREF to the Bayesian network paradigm. The BN-to-URREF mapping is addressed in this paper within the context of the atomic decision procedure (ADP) and the latest view of the URREF ontology. The atomic decision procedure describes the information fusion process from input to output and consists of information sources (ADP-1), the interpretation and processing of this information and the qualification of the uncertainty it contains (ADP-2), the fusion of and reasoning with this uncertain information (ADP-3) and finally the decision scheme and output information (ADP-4). The URREF evaluation of the BN information fusion process allows for evaluation according to (1) representation criteria which relate to ADP-2 and evaluate modeling and model parameterization, (2) reasoning criteria which relate to ADP-3 and evaluate the reasoning scheme, which in the case of a BN entails the computation of marginal probability densities over hypothesis variables and (3) data criteria which relate to ADP-1 and evaluate the sources and the information generated by said sources, together with the qualification and representation of the uncertainty inherent to the infor