共通説明構造の汎化階層上のトップダウン探索による不完全領域理論下での妥当なマクロルールの獲得

概要

論文の詳細を見る
Though EBL (explanation based learning) is a useful framework for generating operational knowledge from an example, we must assume perfect domain theory and explicitly defined operationality criterion. This assumption is too severe in practical situation. This paper presents a topdown search over generalization hierarchy of common explanation structure for solving the imperfect theory problem in EBL, forcusing on an inconsistent problem. Common explanation structures are generated from prural primitive explanation structures and they are organized as a generalization hierarchy based on a concept of maximal covering which reflects on similarity among them. By topdown search over the generalization hierarchy, the best set of macro rules, which includes all positive examples and excludes all negatives, can be found. This method has been implemented on Prolog. Some experiments show its usefulness for valid macro rules under the imperfect domain theory.
社団法人人工知能学会の論文
1991-05-01