No abstract

Verification and validation (V&V) of Knowledge Bases (KBs) are two sides of the same coin: one is intended to assure the structural correctness of the KB, while the other is intended to assure the functional correctness of the domain model embodied in the KB. Knowledge base refinement aims to appropriately revise the KB if a structural or functional error is detected during the V&V process. This paper presents a uniform framework for verification, validation and refinement of KBs represented as sets of production rules, called the VVR system. It incorporates a contradiction-tolerant truth maintenance system (CTMS) for performing both verification and validation analyses, and some simple explanation-based learning techniques for guiding the refinement process. Verification analysis consists of detecting and correcting the main types of structural anomalies: circular rules, redundant rules, inconsistent rules, and inconsistent data, and checks the KB for completeness and violated semantic constraints. In terms of validation, given a set of test cases, the VVR system is capable of detecting and correcting functional errors caused by overgeneralization and/or overspecialization of the KB. If the set of test cases is not available, the VVR system can generate synthetic test cases intended to help the user evaluate KBS performance. 0 1994 John Wiley & Sons, Inc.

In this paper we present work on a project funded by the National Science Foundation with a goal of unifying the Artificial Intelligence (AI) course around the theme of machine learning. Our work involves the development and testing of an adaptable framework for the presentation of core AI topics that emphasizes the relationship between AI and computer science. Several hands-on laboratory projects that can be closely integrated into an introductory AI course have been developed. We present an overview of one of the projects and describe the associated curricular materials that have been developed. The project uses machine learning as a theme to unify core AI topics in the context of the N-puzzle game. Games provide a rich framework to introduce students to search fundamentals and other core AI concepts. The paper presents several pedagogical possibilities for the N-puzzle game, the rich challenge it offers, and summarizes our experiences using it.

Truth maintenance systems (TMSs) were introduced more than ten years ago, but recently there is an explosion of interest in them and their possible applications in different areas. In this paper we discuss truth maintenance from three perspectives: • Truth maintenance as a data base management facility, which was in fact the original intention of the TMS. • Truth maintenance as an infei'ence facility, which provides a way to extend the role of the TMS in solving problems. • Truth maintenance as a verification facility, which illustrates a new and promising application of TMSs in the area of expert systems design. This paper is not intended to provide a complete survey on TMSs, rather it aims to present the basic ideas and functionality of TMS, and to show how different kinds of TMS can be used as a meta-environment for testing Expert System Knowledge Bases, represented as sets of production rules, for anomalies.The paper is addressed to two groups of readers: those who are looking for an introductory survey on TMSs, and those who are interested in non-conventional techniques for Expert System Knowledge Base verification.

No abstract