Assertion generation through active learning

Long H. Pham, Lyly Tran Thi, Jun Sun

Abstract: Program assertions are useful for many program analysis tasks. They are however often missing in practice. In this work, we develop a novel approach for generating likely assertions automatically based on active learning. Our target is complex Java programs which cannot be symbolically executed (yet). Our key idea is to generate candidate assertions based on test cases and then apply active learning techniques to iteratively improve them. The experiments show that active learning really helps to improve the generated assertions.

External IDs:dblp:conf/icse/PhamT017