The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #9840

Title:Implementing Problem Solving Methods in CYC
Date: 1998
Abstract:Although the CYC system is very good in answering questions through backtracking in its large knowledge base (KB), it can be illustrated in the case of systematic fault diagnosis that this is not enough to solve more complex problems that demand dynamic collection of information and KB updating. A solution proposed in this thesis is to provide CYC with a richer set of inference mechanisms and with the means to combine them into more complex problem-solving procedures.The KADS methodology provides an excellent source both for inference mechanisms, it inference types, and ways to combine them, through its inference structures and Generic Task Models. Therefore, KADS provides the means to develop a rich set of inference mechanisms which will improve the reasoning power of CYC. In addition, KADS provides a systematic, task-oriented approach to knowledge acquisition which naturally complements the task-independent development of CYC's Upper Ontology. The combination of these two approaches provides the means to fill in the knowledge gap between there two extremes of the Ontology pyrimid; the task-specific knowlege, which forms the base and the Upper Ontology, which forms the top. A relevent issues discussed, is the problem of "brittleness" of Expert Systems, an issue fundamental for the creation and use of CYC.The actual implementation of the Systematic Diagnosis problem solving method for faults in PCs and Automobiles, described in this thesis, demonstrates a way tom implement KADS problem solving methods in CYC. The implementation maps directly the various layers of KADS Expertise model onto CYC's KB and its LISP-like programming language, SubL. The implemented system thus combines the declarative richness, transparency and expressiveness of the CYC KB with the conceptual analysis and structured search represented in the problem solving method.

[Search These Pages] [DAI Home Page] [Comment]