Abstract
We present a novel approach to efficiently compute tight non-convex enclosures of the image through neural networks with ReLU, sigmoid, or hyperbolic tangent activation functions. In particular, we abstract the input-output relation of each neuron by a polynomial approximation, which is evaluated in a set-based manner using polynomial zonotopes. While our approach can also can be beneficial for open-loop neural network verification, our main application is reachability analysis of neural network controlled systems, where polynomial zonotopes are able to capture the non-convexity caused by the neural network as well as the system dynamics. This results in a superior performance compared to other methods, as we demonstrate on various benchmarks.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
- 2.
References
Althoff, M.: Reachability analysis and its application to the safety assessment of autonomous cars. Ph.D. thesis, Technical University of Munich (2010)
Althoff, M.: Reachability analysis of nonlinear systems using conservative polynomialization and non-convex sets. In: Proceedings of the International Conference on Hybrid Systems: Computation and Control, pp. 173–182 (2013)
Althoff, M.: An introduction to CORA 2015. In: Proceedings of the International Workshop on Applied Verification for Continuous and Hybrid Systems, pp. 120–151 (2015)
Bak, S., Liu, C., Johnson, T.: The second international verification of neural networks competition (VNN-COMP 2021): Summary and results. arXiv:2109.00498 (2021)
Beaufays, F., Abdel-Magid, Y., Widrow, B.: Application of neural networks to load-frequency control in power systems. Neural Netw. 7(1), 183–194 (1994)
Bogomolov, S., et al.: JuliaReach: a toolbox for set-based reachability. In: Proceedings of the International Conference on Hybrid Systems: Computation and Control, pp. 39–44 (2019)
Bunel, R., et al.: Branch and bound for piecewise linear neural network verification. J. Mach. Learn. Res. 21(42) (2020)
Cheng, C.H., Nührenberg, G., Ruess, H.: Maximum resilience of artificial neural networks. In: Proceedings of the International Symposium on Automated Technology for Verification and Analysis, pp. 251–268 (2017)
Christakou, C., Vrettos, S., Stafylopatis, A.: A hybrid movie recommender system based on neural networks. Int. J. Artif. Intell. Tools 16(5), 771–792 (2007)
Clavière, A., et al.: Safety verification of neural network controlled systems. In: Proceedings of the International Conference on Dependable Systems and Networks, pp. 47–54 (2021)
David, O.E., Netanyahu, N.S., Wolf, L.: DeepChess: end-to-end deep neural network for automatic learning in chess. In: Proceedings of the International Conference on Artificial Neural Networks, pp. 88–96 (2016)
Dutta, S., et al.: Learning and verification of feedback control systems using feedforward neural networks. In: Proceedings of the International Conference on Analysis and Design of Hybrid Systems, pp. 151–156 (2018)
Dutta, S., et al.: Sherlock-a tool for verification of neural network feedback systems. In: Proceedings of the International Conference on Hybrid Systems: Computation and Control, pp. 262–263 (2019)
Dutta, S., Chen, X., Sankaranarayanan, S.: Reachability analysis for neural feedback systems using regressive polynomial rule inference. In: Proceedings of the International Conference on Hybrid Systems: Computation and Control, pp. 157–168 (2019)
Ehlers, R.: Formal verification of piece-wise linear feed-forward neural networks. In: Proceedings of the International Symposium on Automated Technology for Verification and Analysis, pp. 269–286 (2017)
Fan, J., Huang, et al.: ReachNN*: a tool for reachability analysis of neural-network controlled systems. In: Proceedings of the International Symposium on Automated Technology for Verification and Analysis, pp. 537–542 (2020)
Goubault, E., Putot, S.: RINO: robust inner and outer approximated reachability of neural networks controlled systems. In: Proceedings of the International Conference on Computer Aided Verification, pp. 511–523 (2022)
Huang, C., et al.: ReachNN: reachability analysis of neural-network controlled systems. Trans. Embed. Comput. Syst. 18(5s) (2019)
Huang, C., et al.: POLAR: A polynomial arithmetic framework for verifying neural-network controlled systems. In: Proceedings of the International Symposium on Automated Technology for Verification and Analysis, pp. 414–430 (2022)
Ivanov, R., et al.: Verisig: verifying safety properties of hybrid systems with neural network controllers. In: Proceedings of the International Conference on Hybrid Systems: Computation and Control, pp. 169–178 (2019)
Ivanov, R., et al.: Verisig 2.0: verification of neural network controllers using Taylor model preconditioning. In: Proceedings of the International Conference on Computer Aided Verification, pp. 249–262 (2021)
Katz, G., et al.: Reluplex: an efficient SMT solver for verifying deep neural networks. In: Proceedings of the International Conference on Computer Aided Verification, pp. 97–117 (2017)
Khan, S., et al.: Facial recognition using convolutional neural networks and implementation on smart glasses. In: Proceedings of the International Conference on Information Science and Communication Technology (2019). Article 19
Khedr, H., Ferlez, J., Shoukry, Y.: PEREGRiNN: penalized-relaxation greedy neural network verifier. In: Proceedings of the International Conference on Computer Aided Verification, pp. 287–300 (2021)
Kochdumper, N., Althoff, M.: Sparse polynomial zonotopes: a novel set representation for reachability analysis. Trans. Autom. Control 66(9), 4043–4058 (2021)
Makino, K., Berz, M.: Taylor models and other validated functional inclusion methods. Int. J. Pure and Appl. Math. 4(4), 379–456 (2003)
Mukherjee, D., et al.: A survey of robot learning strategies for human-robot collaboration in industrial settings. Robot. Comput.-Integr. Manuf. 73 (2022)
Müller, M.N., et al.: PRIMA: precise and general neural network certification via multi-neuron convex relaxations. Proceedings on Programming Languages 1 (2022). Article 43
Müller, M.N., et al.: The third international verification of neural networks competition (VNN-COMP 2022): summary and results. arXiv preprint arXiv:2212.10376 (2022)
Pulina, L., Tacchella, A.: Challenging SMT solvers to verify neural networks. AI Commun. 25(2), 117–135 (2012)
Raghunathan, A., Steinhardt, J., Liang, P.: Semidefinite relaxations for certifying robustness to adversarial examples. In: Proceedings of the International Conference on Neural Information Processing Systems, pp. 10900–10910 (2018)
Riedmiller, M., Montemerlo, M., Dahlkamp, H.: Learning to drive a real car in 20 minutes. In: Proceedings of the International Conference on Frontiers in the Convergence of Bioscience and Information Technologies, pp. 645–650 (2007)
Schilling, C., Forets, M., Guadalupe, S.: Verification of neural-network control systems by integrating Taylor models and zonotopes. In: Proceedings of the AAAI Conference on Artificial Intelligence, pp. 8169–8177 (2022)
Singh, G., et al.: Fast and effective robustness certification. In: Proceedings of the International Conference on Advances in Neural Information Processing Systems (2018)
Singh, G., et al.: An abstract domain for certifying neural networks. Proce. Prog. Lang. 3 (2019). Article 41
Singh, G., et al.: Beyond the single neuron convex barrier for neural network certification. In: Proceedings of the International Conference on Advances in Neural Information Processing Systems (2019)
Tjeng, V., Xiao, K.Y., Tedrake, R.: Evaluating robustness of neural networks with mixed integer programming. In: Proceedings of the International Conference on Learning Representations (2019)
Tran, H.D., et al.: Parallelizable reachability analysis algorithms for feed-forward neural networks. In: Proceedings of the International Conference on Formal Methods in Software Engineering, pp. 51–60 (2019)
Tran, H.D., et al.: Safety verification of cyber-physical systems with reinforcement learning control. Trans. Embedded Comput. Syst. 18(5s) (2019). Article 105
Tran, H.D., et al.: Star-based reachability analysis of deep neural networks. In: Proceedings of the International Symposium on Formal Methods, pp. 670–686 (2019)
Tran, H.D., et al.: NNV: The neural network verification tool for deep neural networks and learning-enabled cyber-physical systems. In: Proc. of the Int. Conf. on Computer Aided Verification. pp. 3–17 (2020)
Vincent, J.A., Schwager, M.: reachable polyhedral marching (RPM): a safety verification algorithm for robotic systems with deep neural network components. In: Proceedings of the International Conference on Robotics and Automation, pp. 9029–9035 (2021)
Wang, S., et al.: Formal security analysis of neural networks using symbolic intervals. In: Proceedings of the USENIX Security Symposium, pp. 1599–1614 (2018)
Wang, S., et al.: Beta-CROWN: efficient bound propagation with per-neuron split constraints for neural network robustness verification. In: Proceedings of the International Conference on Neural Information Processing Systems (2021)
Weng, L., et al.: Towards fast computation of certified robustness for ReLU networks. In: Proceedings of the International Conference on Machine Learning, pp. 5276–5285 (2018)
Xiang, W., et al.: Reachable set estimation and safety verification for piecewise linear systems with neural network controllers. In: Proceedings of the American Control Conference, pp. 1574–1579 (2018)
Xu, K., et al.: Fast and complete: enabling complete neural network verification with rapid and massively parallel incomplete verifiers. In: Proceedings of the International Conference on Learning Representations (2021)
Yang, X., et al.: Reachability analysis of deep ReLU neural networks using facet-vertex incidence. In: Proceedings of the International Conference on Hybrid Systems: Computation and Control (2021). Article 18
Zhang, H., et al.: Efficient neural network robustness certification with general activation functions. In: Proceedings of the International Conference on Neural Information Processing Systems, pp. 4944–4953 (2018)
Acknowledgements
We gratefully acknowledge the financial support from the project justITSELF funded by the European Research Council (ERC) under grant agreement No 817629, from DIREC - Digital Research Centre Denmark, and from the Villum Investigator Grant S4OS. In addition, this material is based upon work supported by the Air Force Office of Scientific Research and the Office of Naval Research under award numbers FA9550-19-1-0288, FA9550-21-1-0121, FA9550-23-1-0066 and N00014-22-1-2156. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the United States Air Force or the United States Navy.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Appendix A
Appendix A
We now provide the proof for Prop. 2. According to Def. 2, the one-dimensional polynomial zonotope \(\mathcal{P}\mathcal{Z}= \langle c,G,G_I,E \rangle _{PZ}\) is defined as
where \(\alpha = [\alpha _1~\dots ~\alpha _p]^T\) and \(\beta = [\beta _1~\dots ~\beta _q]^T\). To compute the image through the quadratic function \(g(x)\) we require the expressions \(d(\alpha )^2\), \(d(\alpha ) z(\beta )\), and \(z(\beta )^2\), which we derive first. For \(d(\alpha )^2\) we obtain
for \(d(\alpha )z(\beta )\) we obtain
and for \(z(\beta )^2\) we obtain
where the function a(i, j) maps indices i, j to a new index:
In (12) and (13), we substituted the expressions \(\beta _j \prod _{k=1}^p \alpha _k^{E_{(k,i)}}\), \(2\beta _i^2 -1\), and \(\beta _i\beta _j\) containing polynomial terms of the independent factors \(\beta \) by new independent factors, which results in an enclosure due to the loss of dependency. The substitution is possible since
Finally, we obtain for the image
which concludes the proof.
Rights and permissions
Copyright information
© 2023 The Author(s), under exclusive license to Springer Nature Switzerland AG
About this paper
Cite this paper
Kochdumper, N., Schilling, C., Althoff, M., Bak, S. (2023). Open- and Closed-Loop Neural Network Verification Using Polynomial Zonotopes. In: Rozier, K.Y., Chaudhuri, S. (eds) NASA Formal Methods. NFM 2023. Lecture Notes in Computer Science, vol 13903. Springer, Cham. https://doi.org/10.1007/978-3-031-33170-1_2
Download citation
DOI: https://doi.org/10.1007/978-3-031-33170-1_2
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-031-33169-5
Online ISBN: 978-3-031-33170-1
eBook Packages: Computer ScienceComputer Science (R0)