Rules for the Use of Dissertations

OF DISSERTATION Inna Valentinovna Pivkina The Graduate School University of Kentucky 2001 REVISION PROGRAMMING: A KNOWLEDGE REPRESENTATION FORMALISM ABSTRACT OF DISSERTATION A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy at the University of KentuckyOF DISSERTATION A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy at the University of Kentucky By Inna Valentinovna Pivkina Lexington, Kentucky Co-Directors: Dr. Wiktor Marek, Professor of Computer Science and Dr. Mirosław Truszczyński, Professor of Computer Science Lexington, Kentucky 2001 ABSTRACT OF DISSERTATIONOF DISSERTATION REVISION PROGRAMMING: A KNOWLEDGE REPRESENTATION FORMALISM The topic of the dissertation is revision programming. It is a knowledge representation formalism for describing constraints on databases, knowledge bases, and belief sets, and providing a computational mechanism to enforce them. Constraints are represented by sets of revision rules. Revision rules could be quite complex and are usually in a form of conditions (for instance, if these elements are present and those elements are absent, then this element must be absent). In addition to being a logical constraint, a revision rule specify a preferred way to satisfy the constraint. Justified revisions semantics assigns to any database a set (possibly empty) of revisions. Each revision satisfies the constraints, and all deletions and additions of elements in a transition from initial database to the revision are derived from revision rules. Revision programming and logic programming are closely related. We established an elegant embedding of revision programs into logic programs, which does not increase the size of a program. Initial database is used in transformation of a revision program into the corresponding logic program, but it is not represented in the logic program. The connection naturally led to extensions of revision programming formalism which correspond to existing extensions of logic programming. More specific, a disjunctive and a nested versions of revision programming were introduced. Also, we studied annotated revision programs, which allow annotations like confidence factors, multiple experts, etc. Annotations were assumed to be elements of a complete infinitely distributive lattice. We proposed a justified revisions semantics for annotated revision programs which agreed with intuitions. Next, we introduced definitions of well-founded semantics for revision programming. It assigns to a revision problem a single “intended” model which is computable in polynomial time. Finally, we extended syntax of revision problems by allowing variables and implemented translators of revision programs into logic programs and a grounder for revision programs. The implementation allows us to compute justified revisions using existing implementations of the stable model semantics for logic programs.