Improving prolog performance by inductive proof generalizations

Abstract

This paper investigates the possibility of improving PROLOG performance by using the history of search for previous queries to guide the search for subsequent queries. First, the notion of a proof plan is introduced, and it is shown how a given logic program can be translated into a plan version of it. The plan version allows search to be guided by a proof plan. A specific plan version called the uninformed plan version allows proofs to be generated from the search performed for any query.

Proof plans can be obtained by a generalization from the collection of proofs obtained earlier. A notion dual to most general unification called most special generalization (msg) is introduced for this purpose. The uniqueness of the msg is shown and an algorithm similar to the unification algorithm is given for computing the msg of two proofs. The soundness of guiding a logic program with a proof plan obtained by such a generalization is then shown.

The soundness breaks down, however, for PROLOG programs that use cut. A method based on multiple plan inference is introduced for such programs and its soundness is proved.

uunetlshaktilbetallcselgv