Inductive Synthesis of Recursive Functional Programs

Abstract. We compare three systems for the task of synthesising functional recursive programs, namely Adate, an approach through evolutionary computation, the classification learner Atre, capable of simultaneously learning mutually dependent, recursive multi-class target predicates, and the inductive/abductive logic program synthesiser DialogsII. An overview over the functionality of all three systems is given and their capabilities are systematically evaluated under equal premises with a variety of recursive problems. We propose to combine Adate’s expressional power with Dialog-II’s search heuristic. The accessory adoption of Atre’s k-beam search strategy to learn mutually dependent recursive target functions should only be adopted, if Adate’s search time could be reduced significantly.

• Type: Conference
• Year: 2007
• Where: KI
• Authors: Martin Hofmann 0008, Andreas Hirschberger, Emanuel Kitzelmann, Ute Schmid
• Keywords: Dependent Recursive Target, Functional Recursive Programs, KI 2007, Recursive Multi-class Target