@inproceedings{d17bf924ba4e4d259cf6bb29a5055279,
title = "Accelerating Safety Verification of Neural Network Dynamical Systems Using Assured Compressed Models",
abstract = "In this paper, we develop a formal assured simplification method to accelerate the safety verification process of a neural network model that approximates a nonlinear dynamical system. Firstly, an assured compression error is proposed to describe the maximum output difference introduced by neural network compression. A reachability-based computational process is then developed to estimate the assured compression error. The assured compression error is used to accelerate the reachability analysis and safety verification of neural network dynamical system models, which can significantly reduce the computational time for safety verification. Finally, the proposed approach is applied to a case study of a magnetic levitation system to demonstrate its correctness and effectiveness.",
keywords = "model compression, neural networks, nonlinear systems, reachable set computation, safety verification",
author = "Zhongzhu Shao and Tao Wang and Weiming Xiang",
note = "Publisher Copyright: {\textcopyright} 2023 IEEE.; 35th Chinese Control and Decision Conference, CCDC 2023 ; Conference date: 20-05-2023 Through 22-05-2023",
year = "2023",
doi = "10.1109/CCDC58219.2023.10327268",
language = "English (US)",
series = "Proceedings of the 35th Chinese Control and Decision Conference, CCDC 2023",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
pages = "4030--4035",
booktitle = "Proceedings of the 35th Chinese Control and Decision Conference, CCDC 2023",
}