Using Previous Experience for Learning Planning Control Knowledge

May 2004

Proceedings of the Seventeenth International Florida Artificial Intelligence Research Society Conference (FLAIRS 2004)

Volume

Proceedings of the Seventeenth International Florida Artificial Intelligence Research Society Conference (FLAIRS 2004)

All Papers

Machine learning (ML) is often used to obtain control knowledge to improve planning efficiency. Usually, ML techniques are used in isolation from experience that could be obtained by other means. The aim of this paper is to determine experimentally the influence of using such previous experience or prior knowledge (PK), so that the learning process is improved. In particular, we study three different ways of getting such experience: from a human, from another planner (called FF), and from a different ML technique. This previous experience has been supplied to two different ML techniques: a deductive-inductive system (hamlet) and a genetic-based one (evock).