Attend and Represent: A Novel View on Algorithm Selection for Software Verification

Today, a plethora of different software verification tools exist. When having a concrete verification task at hand, software developers thus face the problem of algorithm selection. Existing algorithm selectors for software verification typically use handpicked program features together with (1) either manually designed selection heuristics or (2) machine learned strategies. While the first approach suffers from not being transferable to other selection problems, the second approach lacks interpretability, i.e., insights into reasons for choosing particular tools.

In this paper, we propose a novel approach to algorithm selection for software verification. Our approach employs representation learning together with an attention mechanism. Representation learning circumvents feature engineering, i.e., avoids the handpicking of program features. Attention permits a form of interpretability of the learned selectors. We have implemented our approach and have experimentally evaluated and compared it with existing approaches. The evaluation shows that representation learning does not only outperform manual feature engineering, but also enables transferability of the learning model to other selection tasks.