Inductive Synthesis of Recursive Functional Programs A Comparison of Three Systems.pdfVIP

Inductive Synthesis of Recursive Functional Programs A Comparison of Three Systems Martin Hofmann, Andreas Hirschberger, Emanuel Kitzelmannn, and Ute Schmid University of Bamberg, Faculty of Information Systems and Applied Computer Science {martin.hofmann, andreas.hirschberger}@stud.uni-bamberg.de, {emanuel.kitzelmann, ute.schmid}@wiai.uni-bamberg.de 1 Introduction One of the most challenging subfields, and a still little researched niche of ma- chine learning, is the inductive synthesis of recursive programs from incomplete specifications, such as examples for the desired input/output behavior [1–4]. The special appeal of an inductive approach to automated program construction is that the user only provides some examples of the desired program behaviour, such as [A, B, C] → [C, B, A], as input to the synthesis system and a general pro- gram (here for reversing a list) is created. Potential applications for automatic program induction are to enable end-users to create their own simple programs, to assist professional programmers or even to automatically invent new and ef- ficient algorithms. Existing inductive program synthesis systems use either a search-based or an analytical approach. While the most promising are allocated in the subfield of program synthesis, also concept learners with extended codomain and especially ILP-based systems exhibit success worthy to mention. Since no broadly accepted fundamentals and approaches prevail in the field of inductive program synthesis. We have systematically evaluated three systems to inductively synthesise functional recursive programs,1 which are based on three fundamentally different induction methods. 2 The Systems Adate [5] (Automatic Design of Algorithms Through Evolution) is a system for automatic programmin