Betsy.Herk@A.CS.CMU.EDU (03/12/86)
Speaker: Gerald DeJong, University of Illinois Date: Wednesday, April 2 (Note special day/time) Place: 5409 Wean Hall Time: 11:30 - 1:00 Title: Explanation Based Learning Abstract: The schema learning group at Illinois is exploring artificial intelligence techniques that will enable a com- puter system to learn general world knowledge in the form of "schemata" through its interactions with an external environment. A schema is a data structure that specifies, in conceptual terms, a particular real world situation. Schemata can be very useful in problem solving, natural language processing and other AI areas. It is claimed, in this paradigm, that much intelligent behavior can be cap- tured by using a large number of such schemata. The explanation-based method represents a departure from the usual approaches to machine learning in several ways. First, it is very knowledge-based. That is, the sys- tem must possess much knowledge before it can aquire new knowledge. Second, it is capable of one-trial learning. The results so far are promising. Explanation-based learn- ing takes us a large step closer to building an intelligent system capable of learning on its own. A number computer systems have been designed and imple- mented based on Explanatory Schema Acquisition, an explanation-based learning paradigm. The domain areas of these projects include natural language processing, robot- ics, theorem proving, physics problem-solving and theory refinement. Several of the systems will be discussed in the context of theoretical advantages and difficulties with explanation-based learning.