Robustness Analysis of Neural Networks via Efficient Partitioning with Applications in Control Systems
IEEE Neural networks (NNs) are now routinely implemented on systems that must operate in uncertain environments, but the tools for formally analyzing how this uncertainty propagates to NN outputs are not yet commonplace. Computing tight bounds on NN output sets (given an input set) provides a measur...
| Main Authors: | , , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
Institute of Electrical and Electronics Engineers (IEEE)
2021
|
| Online Access: | https://hdl.handle.net/1721.1/134064 |
| _version_ | 1826197841112989696 |
|---|---|
| author | Everett, Michael Habibi, Golnaz How, Jonathan P |
| author2 | Massachusetts Institute of Technology. Aerospace Controls Laboratory |
| author_facet | Massachusetts Institute of Technology. Aerospace Controls Laboratory Everett, Michael Habibi, Golnaz How, Jonathan P |
| author_sort | Everett, Michael |
| collection | MIT |
| description | IEEE Neural networks (NNs) are now routinely implemented on systems that must operate in uncertain environments, but the tools for formally analyzing how this uncertainty propagates to NN outputs are not yet commonplace. Computing tight bounds on NN output sets (given an input set) provides a measure of confidence associated with the NN decisions and is essential to deploy NNs on safety-critical systems. Recent works approximate the propagation of sets through nonlinear activations or partition the uncertainty set to provide a guaranteed outer bound on the set of possible NN outputs. However, the bound looseness causes excessive conservatism and/or the computation is too slow for online analysis. This paper unifies propagation and partition approaches to provide a family of robustness analysis algorithms that give tighter bounds than existing works for the same amount of computation time (or reduced computational effort for a desired accuracy level). Moreover, we provide new partitioning techniques that are aware of their current bound estimates and desired boundary shape (e.g., lower bounds, weighted ℓ∞-ball, convex hull), leading to further improvements in the computation-tightness tradeoff. The paper demonstrates the tighter bounds and reduced conservatism of the proposed robustness analysis framework with examples from model-free RL and forward kinematics learning. |
| first_indexed | 2024-09-23T10:54:07Z |
| format | Article |
| id | mit-1721.1/134064 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T10:54:07Z |
| publishDate | 2021 |
| publisher | Institute of Electrical and Electronics Engineers (IEEE) |
| record_format | dspace |
| spelling | mit-1721.1/1340642023-02-17T19:17:49Z Robustness Analysis of Neural Networks via Efficient Partitioning with Applications in Control Systems Everett, Michael Habibi, Golnaz How, Jonathan P Massachusetts Institute of Technology. Aerospace Controls Laboratory Massachusetts Institute of Technology. Department of Aeronautics and Astronautics IEEE Neural networks (NNs) are now routinely implemented on systems that must operate in uncertain environments, but the tools for formally analyzing how this uncertainty propagates to NN outputs are not yet commonplace. Computing tight bounds on NN output sets (given an input set) provides a measure of confidence associated with the NN decisions and is essential to deploy NNs on safety-critical systems. Recent works approximate the propagation of sets through nonlinear activations or partition the uncertainty set to provide a guaranteed outer bound on the set of possible NN outputs. However, the bound looseness causes excessive conservatism and/or the computation is too slow for online analysis. This paper unifies propagation and partition approaches to provide a family of robustness analysis algorithms that give tighter bounds than existing works for the same amount of computation time (or reduced computational effort for a desired accuracy level). Moreover, we provide new partitioning techniques that are aware of their current bound estimates and desired boundary shape (e.g., lower bounds, weighted ℓ∞-ball, convex hull), leading to further improvements in the computation-tightness tradeoff. The paper demonstrates the tighter bounds and reduced conservatism of the proposed robustness analysis framework with examples from model-free RL and forward kinematics learning. 2021-10-27T19:57:54Z 2021-10-27T19:57:54Z 2021 2021-04-30T15:04:33Z Article http://purl.org/eprint/type/JournalArticle https://hdl.handle.net/1721.1/134064 en 10.1109/LCSYS.2020.3045323 IEEE Control Systems Letters Creative Commons Attribution-Noncommercial-Share Alike http://creativecommons.org/licenses/by-nc-sa/4.0/ application/pdf Institute of Electrical and Electronics Engineers (IEEE) arXiv |
| spellingShingle | Everett, Michael Habibi, Golnaz How, Jonathan P Robustness Analysis of Neural Networks via Efficient Partitioning with Applications in Control Systems |
| title | Robustness Analysis of Neural Networks via Efficient Partitioning with Applications in Control Systems |
| title_full | Robustness Analysis of Neural Networks via Efficient Partitioning with Applications in Control Systems |
| title_fullStr | Robustness Analysis of Neural Networks via Efficient Partitioning with Applications in Control Systems |
| title_full_unstemmed | Robustness Analysis of Neural Networks via Efficient Partitioning with Applications in Control Systems |
| title_short | Robustness Analysis of Neural Networks via Efficient Partitioning with Applications in Control Systems |
| title_sort | robustness analysis of neural networks via efficient partitioning with applications in control systems |
| url | https://hdl.handle.net/1721.1/134064 |
| work_keys_str_mv | AT everettmichael robustnessanalysisofneuralnetworksviaefficientpartitioningwithapplicationsincontrolsystems AT habibigolnaz robustnessanalysisofneuralnetworksviaefficientpartitioningwithapplicationsincontrolsystems AT howjonathanp robustnessanalysisofneuralnetworksviaefficientpartitioningwithapplicationsincontrolsystems |