MVILAIN@G.BBN.COM (Marc Vilain) (04/28/88)
BBN Science Development Program AI Seminar Series Lecture LEARNING EFFECTIVE SEARCH CONTROL KNOWLEDGE: AN EXPLANATION-BASED APPROACH Steven Minton Carnegie-Mellon University (Steven.Minton@cad.cs.cmu.edu) BBN Labs 10 Moulton Street 2nd floor large conference room 10:30 am, Tuesday May 3 In order to solve problems more effectively with accumulating experience, a problem solver must be able to learn and exploit search control knowledge. In this talk, I will discuss the use of explanation-based learning (EBL) for acquiring domain-specific control knowledge. Although previous research has demonstrated that EBL is a viable approach for acquiring control knowledge, in practice EBL may not always generate useful control knowledge. For control knowledge to be effective, the cumulative benefits of applying the knowledge must outweigh the cumulative costs of testing whether the knowledge is applicable. Generating effective control knowledge may be difficult, as evidenced by the complexities often encountered by human knowledge engineers. In general, control knowledge cannot be indiscriminately added to a system; its costs and benefits must be carefully taken into account. To produce effective control knowledge, an explanation-based learner must generate "good" explanations -- explanations that can be profitably employed to control problem solving. In this talk, I will discuss the utility of EBL and describe the PRODIGY system, a problem solver that learns by searching for good explanations. Extensive experiments testing the PRODIGY/EBL architecture in several task domains will be discussed. I will also briefly describe a formal model of EBL and a proof that PRODIGY's generalization algorithm is correct with respect to this model. -------