Apprenticeship Learning in Imperfect Domain Theories

This chapter presents DISCIPLE, a multi-strategy integrated learning s y s t e m illustrating a theory and a methodology for learning expert knowledge in the con tex t of an imperfect domain theory. DISCIPLE integrates a learning system and an e m p t y expert system, both using the same knowledge base. It is initially provided with a n imperfect (nonhomogeneous) domain theory and learns problem solving rules f r o m the problem solving steps received from its expert user, during interactive p r o b l e m solving sessions. In this way, DISCIPLE evolves from a helpful assistant in p r o b l e m solving to a genuine expert. The problem solving method of DISCIPLE c o m b i n e s problem reduction, problem solving by constraints, and problem solving by a n a l o g y . The learning method of DISCIPLE depends of its knowledge about the problem s o l v i n g step (the example) from which it learns. In the context of a complete theory about t h e example, DISCIPLE uses explanation-based learning to improve its performance. I n the context of a weak theory about the example, it synergistically c o m b i n e s explanation-based learning, learning by analogy, empirical learning, and l e a r n i n g by questioning the user, developing its competence. In the context of an i n c o m p l e t e theory about the example, DISCIPLE learns by combining the above m e n t i o n e d methods, improving both its competence and performance. * On leave from CNRS, Univ. Paris-Sud, LRI, Bat. 490, F-91405 Orsay France