RUM: A Layered Architecture for Reasoning with Uncertainty

Abstract

New reasoning techniques for dealing with uncertainty in expert systems have been embedded in RUM, a Reasoning with Uncertainty Module. RUM is an integrated software tool based on a frame system (KEE) that is implemented in an object oriented language. RUM's capabilities are subdivided into three layers: representation, inference, and control. The representation layer is based on frame-like data struc tures that capture the uncertainty information used in the inference layer and the uncertainty mete-information used in the control layer. Linguistic probabilities are used to describe the lower and upper bounds of the certainty mea sure attached to a well formed formula. The source and the conditions under which the information was obtained represent the non-numerical meta-information. The inference layer provides the uncertainty calculi with which to perform the intersection, detachment, union, and pooling of information. Five uncertainty calculi, based on their underlying Triangular norms, are used in this layer. The control layer uses the meta-information to select the appropriate calculus for each context and to resolve even tual ignorance or conflict in the information. This layer enables the programmer to declaratively express the local (context dependent) meta-knowledge that will substitute for the global assumptions traditionally used in uncertain reasoning. RUM has been tested in a sequence of experiments in both naval and aerial situation assessment, consisting of correlating reports and tracks, locating and classifying platforms, and identifying intents and threats. I.