Inductive synthesis of functional programs

Abstract

The introduction of analogy and abstraction as alternative to inductive program synthesis by folding of initial trees is discussed. Analogy is less efficient than folding for generalizing recursive program schemes from finite programs. Using anti-unification, mapping and generalization can be realized within the same framework and it can be demonstrated that abstract program schemes can be learned from some initial programming experience. Moreover, analogy is considered as a crucial strategy in human problem solving and learning. Learning of recursive control rules for artificial intelligence (AI) are also presented as a field of application.