skip to main content
research-article
Public Access

Reachability analysis for neural feedback systems using regressive polynomial rule inference

Published: 16 April 2019 Publication History

Abstract

We present an approach to construct reachable set overapproximations for continuous-time dynamical systems controlled using neural network feedback systems. Feedforward deep neural networks are now widely used as a means for learning control laws through techniques such as reinforcement learning and data-driven predictive control. However, the learning algorithms for these networks do not guarantee correctness properties on the resulting closed-loop systems. Our approach seeks to construct overapproximate reachable sets by integrating a Taylor model-based flowpipe construction scheme for continuous differential equations with an approach that replaces the neural network feedback law for a small subset of inputs by a polynomial mapping. We generate the polynomial mapping using regression from input-output samples. To ensure soundness, we rigorously quantify the gap between the output of the network and that of the polynomial model. We demonstrate the effectiveness of our approach over a suite of benchmark examples ranging from 2 to 17 state variables, comparing our approach with alternative ideas based on range analysis.

References

[1]
Abadi, Martín et al. 2016. TensorFlow: A System for Large-scale Machine Learning. In Proc. OSDI'16. USENIX, 265--283.
[2]
M. Althoff. 2015. An Introduction to CORA 2015. In Proc. of ARCH'15 (EPiC Series in Computer Science), Vol. 34. EasyChair, 120--151.
[3]
S. Bak and M. Caccamo. 2013. Computing Reachability for Nonlinear Systems with HyCreate. In Demo and Poster Session in HSCC'13.
[4]
S. Bak and P. S. Duggirala. 2017. HyLAA: A Tool for Computing Simulation-Equivalent Reachability for Linear Systems. In Proc. of HSCC'17. ACM, 173--178.
[5]
Osbert Bastani, Yani Ioannou, Leonidas Lampropoulos, Dimitrios Vytiniotis, Aditya Nori, and Antonio Criminisi. 2016. Measuring neural net robustness with constraints. In Advances in Neural Information Processing Systems. 2613--2621.
[6]
M. Berz. 1999. Modern Map Methods in Particle Beam Physics. Advances in Imaging and Electron Physics, Vol. 108. Academic Press.
[7]
M. Berz and K. Makino. 1998. Verified Integration of ODEs and Flows Using Differential Algebraic Methods on High-Order Taylor Models. Reliable Computing 4 (1998), 361--369. Issue 4.
[8]
X. Chen. 2015. Reachability Analysis of Non-Linear Hybrid Systems Using Taylor Models. Ph.D. Dissertation. RWTH Aachen University.
[9]
X. Chen, E. Ábrahám, and S. Sankaranarayanan. 2012. Taylor Model Flowpipe Construction for Non-linear Hybrid Systems. In Proc. of RTSS'12. IEEE Computer Society, 183--192.
[10]
X. Chen, E. Ábrahám, and S. Sankaranarayanan. 2013. Flow*: An Analyzer for Non-linear Hybrid Systems. In Proc. of CAV'13 (LNCS), Vol. 8044. Springer, 258--263.
[11]
X. Chen and S. Sankaranarayanan. 2016. Decomposed Reachability Analysis for Nonlinear Systems. In 2016 IEEE Real-Time Systems Symposium (RTSS). IEEE Press, 13--24.
[12]
Antonio Eduardo Carrilho da Cunha. 2015. Benchmark: Quadrotor Attitude Control. In Proc. of ARCH 2015 (EPiC Series in Computing), Vol. 34. EasyChair, 57--72.
[13]
Luiz H. de Figueiredo and Jorge Stolfi. 1997. Self-Validated Numerical Methods and Applications. In Brazilian Mathematics Colloquium monograph. IMPA, Rio de Janeiro, Brazil. Cf. http://www.ic.unicamp.br/~stolfi/EXPORT/papers/by-tag/fig-sto-97-iaaa.ps.gz.
[14]
P. S. Duggirala, S. Mitra, M. Viswanathan, and M. Potok. 2015. C2E2: A Verification Tool for Stateflow Models. In Proc. of TACAS'15 (LNCS), Vol. 9035. Springer, 68--82.
[15]
Souradeep Dutta, Susmit Jha, Sriram Sankaranarayanan, and Ashish Tiwari. 2018. Learning and verification of feedback control systems using feedforward neural networks. IFAC-PapersOnLine 51, 16 (2018), 151--156.
[16]
Souradeep Dutta, Susmit Jha, Sriram Sankaranarayanan, and Ashish Tiwari. 2018. Output Range Analysis for Deep Feedforward Neural Networks. In NASA Formal Methods, Aaron Dutle, César Muñoz, and Anthony Narkawicz (Eds.). Springer International Publishing, Cham, 121--138.
[17]
Rüdiger Ehlers. 2017. Formal Verification of Piece-Wise Linear Feed-Forward Neural Networks. In ATVA (Lecture Notes in Computer Science), Vol. 10482. Springer, 269--286.
[18]
G. Frehse, C. Le Guernic, A. Donzé, S. Cotton, R. Ray, O. Lebeltel, R. Ripado, A. Girard, T. Dang, and O. Maler. 2011. SpaceEx: Scalable Verification of Hybrid Systems. In Proc. of CAV'11 (LNCS), Vol. 6806. Springer, 379--395.
[19]
LiMin Fu. 1994. Rule generation from neural networks. IEEE Transactions on Systems, Man, and Cybernetics 24, 8 (Aug 1994), 1114--1124.
[20]
E. Hainry. 2008. Reachability in Linear Dynamical Systems. In Proc. of CiE 2008 (LNCS), Vol. 5028. Springer, 241--250.
[21]
Xiaowei Huang, Marta Kwiatkowska, Sen Wang, and Min Wu. 2016. Safety Verification of Deep Neural Networks. CoRR abs/1610.06940 (2016). http://arxiv.org/abs/1610.06940
[22]
M. Jankovic, D. Fontaine, and P. V. Kokotovic. 1996. TORA example: cascade- and passivity-based control designs. IEEE Transactions on Control Systems Technology 4, 3 (1996), 292--297.
[23]
Kyle Julian and Mykel J. Kochenderfer. 2017. Neural Network Guidance for UAVs. In AIAA Guidance Navigation and Control Conference (GNC).
[24]
R. R. Kadiyala. 1993. A tool box for approximate linearization of nonlinear systems. IEEE Control Systems 13, 2 (April 1993), 47--57.
[25]
Gregory Kahn, Tianhao Zhang, Sergey Levine, and Pieter Abbeel. 2016. PLATO: Policy Learning using Adaptive Trajectory Optimization. arXiv preprint arXiv:1603.00622 (2016).
[26]
Guy Katz, Clark Barrett, David L. Dill, Kyle Julian, and Mykel J. Kochenderfer. 2017. Reluplex: An Efficient SMT Solver for Verifying Deep Neural Networks. Springer International Publishing, Cham, 97--117.
[27]
S. Kong, S. Gao, W. Chen, and E. M. Clarke. 2015. dReach: Δ-Reachability Analysis for Hybrid Systems. In Proc. of TACAS'15 (LNCS), Vol. 9035. Springer, 200--205.
[28]
Lectures. 2014. Nonlinear Systems and Control. http://people.ee.ethz.ch/~apnoco/Lectures2014/.
[29]
Alessio Lomuscio and Lalit Maganti. 2017. An approach to reachability analysis for feed-forward ReLU neural networks. CoRR abs/1706.07351 (2017). arXiv:1706.07351 http://arxiv.org/abs/1706.07351
[30]
K. Makino and M. Berz. 2003. Taylor models and other validated functional inclusion methods. J. Pure and Applied Mathematics 4, 4 (2003), 379--456.
[31]
S. Mitra and Y. Hayashi. 2000. Neuro-fuzzy rule generation: survey in soft computing framework. IEEE Transactions on Neural Networks 11, 3 (May 2000), 748--768.
[32]
R. E. Moore, R. B. Kearfott, and M. J. Cloud. 2009. Introduction to Interval Analysis. SIAM.
[33]
A. Neumaier. 1993. The Wrapping Effect, Ellipsoid Arithmetic, Stability and Confidence Regions. Springer Vienna, 175--190.
[34]
W. Perruquetti, J. P. Richard, and P. Borne. 1996. Lyapunov analysis ofsliding motions: Application to bounded control. Mathematical Problems in Engineering 3, 1 (1996), 1 -- 25.
[35]
Luca Pulina and Armando Tacchella. 2010. An abstraction-refinement approach to verification of artificial neural networks. In Computer Aided Verification. Springer, 243--257.
[36]
Luca Pulina and Armando Tacchella. 2012. Challenging SMT Solvers to Verify Neural Networks. AI Commun. 25, 2 (2012), 117--135.
[37]
Nathalie Revol and Fabrice Rouillier. 2005. Motivations for an Arbitrary Precision Interval Arithmetic and the MPFI Library. Reliable Computing 11, 4 (2005), 275--290.
[38]
Kazumi Saito and Ryohei Nakano. 2002. Extracting regression rules from neural networks. Neural Networks 15, 10 (2002), 1279 -- 1288.
[39]
Hanan J. Samet. 2006. Foundations of Multidimensional and Metric Data Structures. Morgan Kaufmann.
[40]
Mohamed Amin Ben Sassi, Ezio Bartocci, and Sriram Sankaranarayanan. 2017. A Linear Programming-based iterative approach to Stabilizing Polynomial Dynamics. In Proc. IFAC'17. Elsevier.
[41]
Karsten Scheibler, Leonore Winterer, Ralf Wimmer, and Bernd Becker. 2015. Towards verification of artificial neural networks. In MBMV Workshop. 30âĂŞ40.
[42]
Richard S. Sutton and Andrew G. Barto. 2017. Reinforcement Learning: An Introduction. MIT Press.
[43]
H. Tsukimoto. 2000. Extracting rules from trained neural networks. IEEE Transactions on Neural Networks 11, 2 (Mar 2000), 377--389.
[44]
Weiming Xiang, Hoang-Dung Tran, and Taylor T. Johnson. 2017. Output Reachable Set Estimation and Verification for Multi-Layer Neural Networks. CoRR abs/1708.03322 (2017). arXiv:1708.03322 http://arxiv.org/abs/1708.03322
[45]
Weiming Xiang, Hoang-Dung Tran, and Taylor T. Johnson. 2107. Reachable Set Computation and Safety Verification for Neural Networks with ReLU Activations. Cf. https://arxiv.org/pdf/1712.08163.pdf, posted on ArXIV Dec. 2017.
[46]
Weiming Xiang, Hoang-Dung Tran, Joel A. Rosenfeld, and Taylor T. Johnson. 2018. Reachable Set Estimation and Verification for a Class of Piecewise Linear Systems with Neural Network Controllers. To Appear in the American Control Conference (ACC), invited session on Formal Methods in Controller Synthesis.
[47]
Dong-Hae Yeom and Young Hoon Joo. 2012. Control Lyapunov Function Design by Cancelling Input Singularity. 12 (06 2012).

Cited By

View all
  • (2024)Hybrid Zonotope-Based Backward Reachability Analysis for Neural Feedback Systems With Nonlinear Plant Models2024 American Control Conference (ACC)10.23919/ACC60939.2024.10644573(4155-4161)Online publication date: 10-Jul-2024
  • (2024)PolyARBerNN: A Neural Network Guided Solver and Optimizer for Bounded Polynomial InequalitiesACM Transactions on Embedded Computing Systems10.1145/363297023:2(1-26)Online publication date: 24-Jan-2024
  • (2024)POLAR-Express: Efficient and Precise Formal Reachability Analysis of Neural-Network Controlled SystemsIEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems10.1109/TCAD.2023.333121543:3(994-1007)Online publication date: Mar-2024
  • Show More Cited By

Recommendations

Comments

Information & Contributors

Information

Published In

cover image ACM Conferences
HSCC '19: Proceedings of the 22nd ACM International Conference on Hybrid Systems: Computation and Control
April 2019
299 pages
ISBN:9781450362825
DOI:10.1145/3302504
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions from [email protected]

Sponsors

Publisher

Association for Computing Machinery

New York, NY, United States

Publication History

Published: 16 April 2019

Permissions

Request permissions for this article.

Check for updates

Author Tags

  1. flowpipe construction
  2. hybrid system
  3. neural network
  4. polynomial regression
  5. reachability analysis

Qualifiers

  • Research-article

Funding Sources

Conference

HSCC '19
Sponsor:

Acceptance Rates

Overall Acceptance Rate 153 of 373 submissions, 41%

Contributors

Other Metrics

Bibliometrics & Citations

Bibliometrics

Article Metrics

  • Downloads (Last 12 months)437
  • Downloads (Last 6 weeks)51
Reflects downloads up to 22 Oct 2024

Other Metrics

Citations

Cited By

View all
  • (2024)Hybrid Zonotope-Based Backward Reachability Analysis for Neural Feedback Systems With Nonlinear Plant Models2024 American Control Conference (ACC)10.23919/ACC60939.2024.10644573(4155-4161)Online publication date: 10-Jul-2024
  • (2024)PolyARBerNN: A Neural Network Guided Solver and Optimizer for Bounded Polynomial InequalitiesACM Transactions on Embedded Computing Systems10.1145/363297023:2(1-26)Online publication date: 24-Jan-2024
  • (2024)POLAR-Express: Efficient and Precise Formal Reachability Analysis of Neural-Network Controlled SystemsIEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems10.1109/TCAD.2023.333121543:3(994-1007)Online publication date: Mar-2024
  • (2024)Reachability Verification Based Reliability Assessment for Deep Reinforcement Learning Controlled Robotics and Autonomous SystemsIEEE Robotics and Automation Letters10.1109/LRA.2024.33644719:4(3299-3306)Online publication date: Apr-2024
  • (2024)Collision Avoidance Verification of Multiagent Systems With Learned PoliciesIEEE Control Systems Letters10.1109/LCSYS.2024.34001908(652-657)Online publication date: 2024
  • (2024)Online Data-Driven Safety Certification for Systems Subject to Unknown Disturbances2024 IEEE International Conference on Robotics and Automation (ICRA)10.1109/ICRA57147.2024.10610163(9939-9945)Online publication date: 13-May-2024
  • (2024)Smooth compositions enhance safety of the fuzzy systemsFuzzy Sets and Systems10.1016/j.fss.2024.108888484:COnline publication date: 15-May-2024
  • (2024)Verifying the Generalization of Deep Learning to Out-of-Distribution DomainsJournal of Automated Reasoning10.1007/s10817-024-09704-768:3Online publication date: 3-Aug-2024
  • (2024)Case Study: Runtime Safety Verification of Neural Network Controlled SystemRuntime Verification10.1007/978-3-031-74234-7_13(205-217)Online publication date: 12-Oct-2024
  • (2023)Safety verification of decision-tree policies in continuous timeProceedings of the 37th International Conference on Neural Information Processing Systems10.5555/3666122.3666772(14750-14769)Online publication date: 10-Dec-2023
  • Show More Cited By

View Options

View options

PDF

View or Download as a PDF file.

PDF

eReader

View online with eReader.

eReader

Get Access

Login options

Media

Figures

Other

Tables

Share

Share

Share this Publication link

Share on social media