Abstract
Local robustness refers to whether a deep neural network (DNN) can correctly classify an image under certain perturbations (e.g. Gaussian noise and the \(L_{\infty }\) perturbation). Since the discovery of adversarial examples, the local robustness of DNNs has received much attention, and researchers have proposed many formal verification techniques to measure it. One important application of these verification techniques is to compare the local robustness of different DNNs. However, these techniques contain some parameters that need to be manually set, and it is unclear whether the selection of parameters will affect the comparison results. In this paper, we conduct an empirical study to explore DNNs’ local robustness towards perturbations. We find that two widely-used assumptions in existing papers are not always true in practice. Based on our experimental results, we discuss defects of existing local robustness comparison methods and provide some possible solutions.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
References
Anderson, B.G., Sojoudi, S.: Certifying neural network robustness to random input noise from samples (2020). arXiv:2010.07532
Baluta, T., Chua, Z.L., Meel, K.S., Saxena, P.: Scalable quantitative verification for deep neural networks. In: 43th International Conference on Software Engineering, pp. 312–323 (2021)
Bu, H., Sun, M.: Certifying semantic robustness of deep neural networks. In: 27th International Conference on Engineering of Complex Computer Systems, pp. 51–60 (2023)
Bu, H., Sun, M.: Measuring robustness of deep neural networks from the lens of statistical model checking. In: 2023 International Joint Conference on Neural Networks (2023)
Cardelli, L., Kwiatkowska, M., Laurenti, L., Paoletti, N., Patane, A., Wicker, M.: Statistical guarantees for the robustness of Bayesian neural networks. In: 28th International Joint Conference on Artificial Intelligence, pp. 5693–5700 (2019)
Hendrycks, D., Dietterich, T.: Benchmarking neural network robustness to common corruptions and perturbations. In: International Conference on Learning Representations (2019)
Huang, C., Hu, Z., Huang, X., Pei, K.: Statistical certification of acceptable robustness for neural networks. In: 30th International Conference on Artificial Neural Networks, pp. 79–90 (2021)
Huang, X., Kwiatkowska, M., Wang, S., Wu, M.: Safety verification of deep neural networks. In: 29th International Conference on Computer Aided Verification, pp. 3–29 (2017)
Katz, G., Barrett, C., Dill, D.L., Julian, K., Kochenderfer, M.J.: Reluplex: an efficient SMT solver for verifying deep neural networks. In: 29th International Conference on Computer Aided Verification, pp. 97–117 (2017)
Levy, N., Katz, G.: RoMA: a method for neural network robustness measurement and assessment. In: 29th International Conference on Neural Information Processing (2022)
Li, R., Yang, P., Huang, C.C., Sun, Y., Xue, B., Zhang, L.: Towards practical robustness analysis for DNNs based on PAC-model learning. In: 44th International Conference on Software Engineering, pp. 2189–2201 (2022)
Lin, W., et al.: Robustness verification of classification deep neural networks via linear programming. In: IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 11418–11427 (2019)
Mohapatra, J., Weng, T.W., Chen, P.Y., Liu, S., Daniel, L.: Towards verifying robustness of neural networks against a family of semantic perturbations. In: IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 244–252 (2020)
Pautov, M., Tursynbek, N., Munkhoeva, M., Muravev, N., Petiushko, A., Oseledets, I.: CC-Cert: a probabilistic approach to certify general robustness of neural networks. In: AAAI Conference on Artificial Intelligence (2022)
Singh, G., Gehr, T., Mirman, M., Püschel, M., Vechev, M.: Fast and effective robustness certification. In: Proceedings of the Advances in Neural Information Processing Systems (2018)
Tit, K., Furon, T., Rousset, M.: Efficient statistical assessment of neural network corruption robustness. In: Proceedings of the Advances in Neural Information Processing Systems (2021)
Webb, S., Rainforth, T., Teh, Y.W., Kumar, M.P.: A statistical approach to assessing neural network robustness. In: International Conference on Learning Representations (2019)
Zhang, H., Weng, T.W., Chen, P.Y., Hsieh, C.J., Daniel, L.: Efficient neural network robustness certification with general activation functions. In: Proceedings of the Advances in Neural Information Processing Systems (2018)
Acknowledgements
This work was sponsored by the National Natural Science Foundation of China under grant No. 62172019 and CCF-Huawei Populus Grove Fund.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2023 The Author(s), under exclusive license to Springer Nature Switzerland AG
About this paper
Cite this paper
Bu, H., Sun, M. (2023). Guiding the Comparison of Neural Network Local Robustness: An Empirical Study. In: Iliadis, L., Papaleonidas, A., Angelov, P., Jayne, C. (eds) Artificial Neural Networks and Machine Learning – ICANN 2023. ICANN 2023. Lecture Notes in Computer Science, vol 14258. Springer, Cham. https://doi.org/10.1007/978-3-031-44192-9_25
Download citation
DOI: https://doi.org/10.1007/978-3-031-44192-9_25
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-031-44191-2
Online ISBN: 978-3-031-44192-9
eBook Packages: Computer ScienceComputer Science (R0)