Learning Goal-Decomposition Rules Using Exercises

Abstract

Exercises are problems ordered in increasing order of difficulty. Teaching problemsolving through exercises is a widely used pedagogic technique. A computational reason for this is that the knowledge gained by solving simple problems is useful in efficiently solving more difficult problems. We adopt this approach of learning from exercises to acquire search-control knowledge in the form of goal-decomposition rules (d-rules). Drules are first order, and are learned using a new "generalize-and-test" algorithm which is based on inductive logic programming techniques. We demonstrate the feasibility of the approach by applying it in two planning domains.