A unified representation of uncertainty
Neal T Christensen · 1986
This work focuses on the problem of representing uncertainty in automated reasoning systems such as expert systems. Our goal is to adopt an approach in which the representation of uncertainty is uniform regardless of the source of uncertainty, that produces results in which the user of the system will have more confidence, and with which it is easily possible to explain results. Most significantly, we demonstrate that both uncertainty specification and uncertainty manipulation can be defined at the time the inference system is defined rather than at the time of application. Original work presented here includes the concept of fundamental, or 'atomic', certainties determined at the time of system definition, eliminating the need for assigning uncertainty values at the time of application; the axiomatic development of a method of information fusion; explicit specification of negations; and a proposal for the evaluation of hypotheses based on a 'certainty' ordering. The ideas developed are demonstrated with an example, and results are provided through the use of a lisp implementation of the model and data bases for several situations.