UTCS publications最新文献

{"title":"Learning as knowledge integration","authors":"K. Murray","doi":"10.15781/T2862BX6H","DOIUrl":"https://doi.org/10.15781/T2862BX6H","url":null,"abstract":"A fundamental challenge for Artificial Intelligence is developing methods to build and maintain knowledge-based systems. Knowledge integration is the task of identifying how new and prior knowledge interact while incorporating new information into a knowledge base. This task is pervasive because substantial knowledge bases must be developed incrementally: segments of knowledge are added separately to a growing body of knowledge. This task is difficult because new and prior knowledge may interact in very subtle and surprising ways, and unanticipated interactions may require changes to the knowledge base. Performing knowledge integration involves determining and effecting these changes. This research investigates knowledge integration as a machine learning task. Its contributions include formalizing knowledge integration as a machine learning task, developing a computational model for performing knowledge integration, and instantiating the computational model as an implemented machine learning program. The study of knowledge integration and methods that perform it is important both for pragmatic concerns of building knowledge-based systems and for theoretical concerns of understanding learning systems. By identifying subtle conflicts and gaps in knowledge, knowledge integration facilitates building knowledge-based systems. By avoiding unnecessary restrictions on learning situations, knowledge integration reveals important sources of learning bias and permits learning behaviors that are more opportunistic than do traditional machine learning tasks. REACT is a computational model that identifies three essential activities for performing knowledge integration. Elaboration assesses how new and prior knowledge interact. The system''s limited capacity to explore the interactions of new and prior knowledge requires methods to focus its attention. This focus is achieved by restricting elaboration to consider only selected segments of prior knowledge. Recognition selects the prior knowledge that is considered during elaboration. By identifying the consequences of new information for relevant prior knowledge, recognition and elaboration reveal learning opportunities, such as inconsistencies and gaps in the extended knowledge base. Adaptation exploits these learning opportunities by modifying the new or prior knowledge. KI is a machine learning program that implements the REACT model. Empirical studies demonstrate that KI provides significant assistance to knowledge engineers while integrating new information into a large knowledge base.","PeriodicalId":159150,"journal":{"name":"UTCS publications","volume":"196 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1995-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132683891","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}