Proceedings:
Book One
Volume
Issue:
Proceedings of the AAAI Conference on Artificial Intelligence, 6
Track:
Machine Learning and Knowledge Acquisition
Downloads:
Abstract:
Explanation-Based Generalization (EBG) has been recently a much-explored method of generalization. By utilizing domain knowledge, and knowledge of the concept being learned, EBG produced a valid generalization from a single example. Most EBG systems are currently provided with the concept being learned-or target concept-as a fixed input. A more robust generalization mechanism needs the ability to automatically formulate appropriate target concepts based on the purpose of the learning, since concepts learned for one purpose may not be appropriate for another. This paper introduced a technique and an implemented system that automatically formulate target concepts and their specialized definitions. In particular, the technique derives definitions of everyday artifacts (e.g. CUP), from information about the purpose for which agents intend to use them (e.g. to satisfy their thirst). Given two different purposes for which an agent might use a cup (e.g. as an ornament, versus to satisfy thirst), two different definitions can be derived.
AAAI
Proceedings of the AAAI Conference on Artificial Intelligence, 6