The paper presents an interesting contribution to deep nets verification. While the paper tackles a significant problem, a better reference to previous work will certainly improve this submission. Particularly, it will be nice to emphasize the importance of the abstraction, both practically and theoretically.