{"title":"A knowledge-level analysis of explanation-based learning","authors":"M. Numao, M. Shimura","doi":"10.1145/98894.99105","DOIUrl":null,"url":null,"abstract":"Although Explanation-Based Learning (EBL) has up to now been used only for deductive learning that improves execution speed, chunking in Soar, which is closely related to EBL, was demonstrated to acquire new knowledge. We first analyze such knowledge level learning in EBL, by showing that a rule set is specialized when rules in it are replaced by their composition, and is generalized when a rule is replaced by its decomposition. Counting on this discussion, we propose a method to learn generalized rules by making a decomposition of instances. Since this method acquires knowledge that is deduced from domain theory and induced from instances, it is a natural method for combining empirical and explanation-based learning. We demonstrate deductive and inductive aspects of our method by examples of logic circuit design and geometric analogy.","PeriodicalId":175812,"journal":{"name":"Proceedings of the 3rd international conference on Industrial and engineering applications of artificial intelligence and expert systems - Volume 2","volume":"20 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"1990-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"3","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Proceedings of the 3rd international conference on Industrial and engineering applications of artificial intelligence and expert systems - Volume 2","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1145/98894.99105","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 3
Abstract
Although Explanation-Based Learning (EBL) has up to now been used only for deductive learning that improves execution speed, chunking in Soar, which is closely related to EBL, was demonstrated to acquire new knowledge. We first analyze such knowledge level learning in EBL, by showing that a rule set is specialized when rules in it are replaced by their composition, and is generalized when a rule is replaced by its decomposition. Counting on this discussion, we propose a method to learn generalized rules by making a decomposition of instances. Since this method acquires knowledge that is deduced from domain theory and induced from instances, it is a natural method for combining empirical and explanation-based learning. We demonstrate deductive and inductive aspects of our method by examples of logic circuit design and geometric analogy.