Inductive Program Synthesis: From Theory to Application

We present an approach to folding of finite program terms based on the detection of recurrence relations in a single given term which is considered as the kth unfolding of an unknown recursive program. Our approach goes beyond Summers’ classical approach of inductive program synthesis in several aspects and allows to deal with a larger class of programming problems. We show how inductive synthesis of recursive programs can be applied to control-rule learning and enduser programming: finite program terms are generated using domain-specific techniques and then presented as input to our folding algorithm. In the case of control-rule learning, the finite terms are obtained from a state-based planner and the resulting recursive functions represent knowledge how to guide plan construction. In the case of enduser programming, the user presents examples of the desired input/output relation, and a finite program, which transforms exactly the given IO pairs, is constructed by generate-and-test. The folded recursive program generalizes over the input domain and therefore can be applied to a larger class of problems.