@inproceedings{c3dc7aef6a5245f4b9c94bcc5d0c6982,
title = "An Abstraction Neural Network Generator for Efficient Formal Verification",
abstract = "Deep neural networks have been increasingly used in safety-critical systems as an important component. Most methods of verifying neural networks are difficult to apply to large neural networks because of the complexity of the neural network. Abstraction has proven to be an effective way to improve scalability. A challenging problem is that the output precision is negatively correlated with the degree of abstraction. In this paper, we propose a Distance-Based Merging (DBM) operator to optimize the abstract operation. Also, we propose a Dynamic-Merging Abstraction (DMA) algorithm to heuristically merge the network. Experimental results show that our method outperforms other state-of-the-art methods in terms of output precision.",
keywords = "Abstraction, Neural Network, Over Approximation",
author = "Shengkai Xu and Min Zhang and Xiaodong Zheng and Zhaohui Wang and Bojie Shao",
note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd 2023.; 3rd International Conference on Artificial Intelligence Logic and Applications, AILA 2023 ; Conference date: 05-08-2023 Through 06-08-2023",
year = "2023",
doi = "10.1007/978-981-99-7869-4\_11",
language = "英语",
isbn = "9789819978687",
series = "Communications in Computer and Information Science",
publisher = "Springer Science and Business Media Deutschland GmbH",
pages = "139--152",
editor = "Songmao Zhang and Yonggang Zhang",
booktitle = "Artificial Intelligence Logic and Applications - The 3rd International Conference, AILA 2023, Proceedings",
address = "德国",
}