Speed - up Transformations of Logic

This paper presents a formal framework for the automatic transformation of ineecient generate and test programs into equivalent programs by learning from the proofs of example queries. In the resulting program the search is guided by strategies based on abstracted proof traces obtained from the interpretation of example computations. Strategies are incre-mentally improved in an iterative process by a method for theory formation. For the task of theory formation we develop a logic based method. It operates on a triple, consisting of a set of features augmented with taxonomic relations, a sequence of positive and negative facts and a set of disjunctive hypotheses, which can also be the empty set. This triple is transformed into a set of disjunctive hypotheses which implies all of the positive facts, but none of the negative facts and the number of hypotheses is minimal. Hypotheses are constructed from facts by abstraction.