Apprenticeship learning of query based problem solving rules

Abstract

The authors propose LASSP, an apprentice system for learning problem-solving rules in domains where solving a problem entails multiple, possibly dependent, queries performed on an underlying database. Contrary to other learning apprentice systems, LASSP has a domain theory that is initially empty, yet ever improves in terms of its strength and correctness as successive problem-solving episodes are experienced. LASSP has the desirable property of being able to learn and exploit its domain theory concurrently. Foremost, LASSP can asynchronously integrate expertise of multiple experts into a common domain theory, in a seamless manner. This parallel assimilation of knowledge potentially enables the apprentice to develop its domain theory in a fraction of the time normally needed in serial assimilation approaches.<>