Structural Coverage Criteria for Neural Networks Could Be MisleadingNIER
There is a dramatically increasing interest in the quality assurance for DNN-based systems in the software engineering community. An emerging hot topic in this direction is structural coverage criteria for testing neural networks, which are inspired by coverage metrics used in conventional software testing. In this short paper, we argue that these criteria could be misleading because of the fundamental differences between neural networks and human written programs. Our preliminary exploration shows that (1) adversarial examples are pervasively distributed in the finely divided space defined by such coverage criteria, while available natural samples are very sparse, and as a consequence, (2) previously reported fault-detection “capabilities” conjectured from high coverage testing are more likely due to the adversary-oriented search but not the real “high” coverage.
Fri 31 May Times are displayed in time zone: Eastern Time (US & Canada) change
14:00 - 15:30: Testing of AI SystemsPapers / New Ideas and Emerging Results / Demonstrations / Technical Track at Place du Canada Chair(s): Marija MikicGoogle | |||
14:00 - 14:20 Talk | CRADLE: Cross-Backend Validation to Detect and Localize Bugs in Deep Learning LibrariesTechnical Track Technical Track Viet Hung PhamUniversity of Waterloo, Thibaud Lutellier, Weizhen QiUniversity of Science and Technology of China, Lin TanPurdue University Pre-print | ||
14:20 - 14:40 Talk | Guiding Deep Learning System Testing using Surprise Adequacy Technical Track Jinhan KimKAIST, Robert FeldtChalmers University of Technology, Shin YooKorea Advanced Institute of Science and Technology Authorizer link Pre-print | ||
14:40 - 15:00 Talk | DeepConcolic: Testing and Debugging Deep Neural NetworksDemos Demonstrations Youcheng SunUniversity of Oxford, Xiaowei HuangUniversity of Liverpool, Daniel KroeningUniversity of Oxford, James SharpDefence Science and Technology Laboratory (Dstl), Matthew HillDefence Science and Technology Laboratory (Dstl), Rob AshmoreDefence Science and Technology Laboratory (Dstl) | ||
15:00 - 15:10 Talk | Towards Improved Testing For Deep LearningNIER New Ideas and Emerging Results Pre-print | ||
15:10 - 15:20 Talk | Structural Coverage Criteria for Neural Networks Could Be MisleadingNIER New Ideas and Emerging Results Zenan LiNanjing University, Xiaoxing MaNanjing University, Chang XuNanjing University, Chun CaoNanjing University Pre-print | ||
15:20 - 15:30 Talk | Robustness of Neural Networks: A Probabilistic and Practical PerspectiveNIER New Ideas and Emerging Results |