Improving the Representation Space through Exception-Based Learning

This paper addresses the problem of improving the representation space in a rule-based intelligent system, through exception-based learning. Such a system generally learns rules containing exceptions because its representation language is incomplete. However, these exceptions suggest what may be missing from the system's ontology, which is the basis of the representation language. We describe an interactive exception-based learning method for eliciting new elements in the system's ontology in order to eliminate the exceptions of the rules. This method is implemented in the Disciple learning agent shell and has been evaluated in an agent training experiment at the US Army War College.

• Type: Conference
• Year: 2003
• Where: FLAIRS
• Authors: Cristina Boicu, Gheorghe Tecuci, Mihai Boicu, Dorin Marcu
• Keywords: Artificial Intelligence, Exception-based Learning, FLAIRS 2003, Representation Language, System's Ontology