Bounding the cost of learned rules: A transformational approach

Cost reduction of explanation-based learning (EBL) systems is studied using transformational analysis. In this method, the learning process is decomposed into a sequence of transformations from the problem solving to the learned rules. The structure of the problem solving is varied from the match process from learned rules; and search control rules and the optimization employed in the problem solving are ignored. Results on a set of known expensive-rule learning tasks show that such modifications can effectively eliminate the identified set of sources of expensiveness.