Font Size: a A A

Learning horn-clauses as classification rules for relations

Posted on:1993-09-13Degree:Ph.DType:Dissertation
University:Louisiana State University and Agricultural & Mechanical CollegeCandidate:Langley, Mary PamelaFull Text:PDF
GTID:1478390014997410Subject:Computer Science
Abstract/Summary:
Explanation-based learning (EBL) has been applied numerous times in many different domains using as much background knowledge as possible to guide the learning process. We design and implement a new EBL based algorithm called FORGE (FOrming Rules from Ground Explanations) which operates utilizing limited background knowledge. The input to the FORGE algorithm consists of examples and counterexamples of the target concept in the form of ordered tuples along with example tuples of base relations. This limited input is used to construct explanation trees which produce ground rules. These ground rules are then generalized and evaluated using simple cover counts. Different heuristics for pruning the explanation trees are proposed and examined. The FORGE algorithm is compared to the empirically based FOIL algorithm and it is shown that the number of rules evaluated is often significantly reduced by utilizing the EBL methods of FORGE even in the limited knowledge based domain of FOIL.
Keywords/Search Tags:EBL, FORGE, Rules
Related items