Knowledge Representation and Inference in Knowledge Based Systems (Expert Systems)

Abstract

Knowledge Based Systems (KBS) are computer programs in which knowledge and control arc explicitly separated. In first generation KBS, the reasoning is usually monotonic and the control is procedural. Second generation KBS usually exhibit nonmonotonic reasoning, declarative control, and more sophisticated representations of uncertainty. We will focus on the first generation KBS and analyze their typical architecture, composed of a Knowledge Base (KB), a Working Memory (WM), and an Inference Engine (IE). The KB describes the domain knowledge; the WM describes a problem instance; the IE determines the applicability of different subsets of the KB to the current problem. The selection of a specific knowledge representation paradigm, used to build the KB, implicitly determines the selection of the inference mechanism to be used. We will briefly discuss Predicate Calculus, which uses unification and resolution, Frames, which use inheritance, and Production Rules, which use rule chaining.