Programming by demonstration

Abstract

Although Programming by Demonstration (PBD) has the potential to improve the productivity of unsophis-ticated users, previous PBD systems have used brittle, heuristic, domain-speci c approaches to execution-trace generalization. In this paper we de ne two application-independent methods for performing generalization that are based on well-understood machine learning technol-ogy.,vs uses version-space generalization, and,foil is based on the FOIL inductive logic pro-gramming algorithm. We analyze each method both theoretically and empirically, arguing that TGen vs has lower sample complexity, but TGen foil can learn a much more interesting class of programs.