An RNN-Based Framework for the MILP Problem in Robustness Verification of Neural Networks

Hao Xue; Xia Zeng; Wang Lin; Zhengfeng Yang; Chao Peng; Zhenbing Zeng

doi:10.1007/978-3-031-26319-4_34

An RNN-Based Framework for the MILP Problem in Robustness Verification of Neural Networks

Hao Xue
, Xia Zeng
, Wang Lin
, Zhengfeng Yang^*
, Chao Peng
, Zhenbing Zeng

^*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

Robustness verification of ‘s is becoming increasingly crucial for their potential use in many safety-critical applications. Essentially, the problem of robustness verification can be encoded as a typical Mixed-Integer Linear Programming (MILP) problem, which can be solved via branch-and-bound strategies. However, these methods can only afford limited scalability and remain challenging for verifying large-scale neural networks. In this paper, we present a novel framework to speed up the solving of the MILP problems generated from the robustness verification of deep neural networks. It employs a semi-planet relaxation to abstract ReLU activation functions, via an RNN-based strategy for selecting the relaxed ReLU neurons to be tightened. We have developed a prototype tool L2T and conducted comparison experiments with state-of-the-art verifiers on a set of large-scale benchmarks. The experiments show that our framework is both efficient and scalable even when applied to verify the robustness of large-scale neural networks.

Original language	English
Title of host publication	Computer Vision – ACCV 2022 - 16th Asian Conference on Computer Vision, 2022, Proceedings
Editors	Lei Wang, Juergen Gall, Tat-Jun Chin, Imari Sato, Rama Chellappa
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	571-586
Number of pages	16
ISBN (Print)	9783031263187
DOIs	https://doi.org/10.1007/978-3-031-26319-4_34
State	Published - 2023
Event	16th Asian Conference on Computer Vision, ACCV 2022 - Macao, China Duration: 4 Dec 2022 → 8 Dec 2022

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	13841 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	16th Asian Conference on Computer Vision, ACCV 2022
Country/Territory	China
City	Macao
Period	4/12/22 → 8/12/22

Keywords

Learning methods
Neural networks
Robustness verification
Semi-planet relaxation

Access to Document

10.1007/978-3-031-26319-4_34

Cite this

Xue, H., Zeng, X., Lin, W., Yang, Z., Peng, C., & Zeng, Z. (2023). An RNN-Based Framework for the MILP Problem in Robustness Verification of Neural Networks. In L. Wang, J. Gall, T.-J. Chin, I. Sato, & R. Chellappa (Eds.), Computer Vision – ACCV 2022 - 16th Asian Conference on Computer Vision, 2022, Proceedings (pp. 571-586). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13841 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-031-26319-4_34

Xue, Hao ; Zeng, Xia ; Lin, Wang et al. / An RNN-Based Framework for the MILP Problem in Robustness Verification of Neural Networks. Computer Vision – ACCV 2022 - 16th Asian Conference on Computer Vision, 2022, Proceedings. editor / Lei Wang ; Juergen Gall ; Tat-Jun Chin ; Imari Sato ; Rama Chellappa. Springer Science and Business Media Deutschland GmbH, 2023. pp. 571-586 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{0fab0cf5b4e640e6939a64af8c05ebf0,

title = "An RNN-Based Framework for the MILP Problem in Robustness Verification of Neural Networks",

abstract = "Robustness verification of {\textquoteleft}s is becoming increasingly crucial for their potential use in many safety-critical applications. Essentially, the problem of robustness verification can be encoded as a typical Mixed-Integer Linear Programming (MILP) problem, which can be solved via branch-and-bound strategies. However, these methods can only afford limited scalability and remain challenging for verifying large-scale neural networks. In this paper, we present a novel framework to speed up the solving of the MILP problems generated from the robustness verification of deep neural networks. It employs a semi-planet relaxation to abstract ReLU activation functions, via an RNN-based strategy for selecting the relaxed ReLU neurons to be tightened. We have developed a prototype tool L2T and conducted comparison experiments with state-of-the-art verifiers on a set of large-scale benchmarks. The experiments show that our framework is both efficient and scalable even when applied to verify the robustness of large-scale neural networks.",

keywords = "Learning methods, Neural networks, Robustness verification, Semi-planet relaxation",

author = "Hao Xue and Xia Zeng and Wang Lin and Zhengfeng Yang and Chao Peng and Zhenbing Zeng",

note = "Publisher Copyright: {\textcopyright} 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.; 16th Asian Conference on Computer Vision, ACCV 2022 ; Conference date: 04-12-2022 Through 08-12-2022",

year = "2023",

doi = "10.1007/978-3-031-26319-4\_34",

language = "英语",

isbn = "9783031263187",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "571--586",

editor = "Lei Wang and Juergen Gall and Tat-Jun Chin and Imari Sato and Rama Chellappa",

booktitle = "Computer Vision – ACCV 2022 - 16th Asian Conference on Computer Vision, 2022, Proceedings",

address = "德国",

}

Xue, H, Zeng, X, Lin, W, Yang, Z , Peng, C & Zeng, Z 2023, An RNN-Based Framework for the MILP Problem in Robustness Verification of Neural Networks. in L Wang, J Gall, T-J Chin, I Sato & R Chellappa (eds), Computer Vision – ACCV 2022 - 16th Asian Conference on Computer Vision, 2022, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 13841 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 571-586, 16th Asian Conference on Computer Vision, ACCV 2022, Macao, China, 4/12/22. https://doi.org/10.1007/978-3-031-26319-4_34

An RNN-Based Framework for the MILP Problem in Robustness Verification of Neural Networks. / Xue, Hao; Zeng, Xia; Lin, Wang et al.
Computer Vision – ACCV 2022 - 16th Asian Conference on Computer Vision, 2022, Proceedings. ed. / Lei Wang; Juergen Gall; Tat-Jun Chin; Imari Sato; Rama Chellappa. Springer Science and Business Media Deutschland GmbH, 2023. p. 571-586 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13841 LNCS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - An RNN-Based Framework for the MILP Problem in Robustness Verification of Neural Networks

AU - Xue, Hao

AU - Zeng, Xia

AU - Lin, Wang

AU - Yang, Zhengfeng

AU - Peng, Chao

AU - Zeng, Zhenbing

PY - 2023

Y1 - 2023

N2 - Robustness verification of ‘s is becoming increasingly crucial for their potential use in many safety-critical applications. Essentially, the problem of robustness verification can be encoded as a typical Mixed-Integer Linear Programming (MILP) problem, which can be solved via branch-and-bound strategies. However, these methods can only afford limited scalability and remain challenging for verifying large-scale neural networks. In this paper, we present a novel framework to speed up the solving of the MILP problems generated from the robustness verification of deep neural networks. It employs a semi-planet relaxation to abstract ReLU activation functions, via an RNN-based strategy for selecting the relaxed ReLU neurons to be tightened. We have developed a prototype tool L2T and conducted comparison experiments with state-of-the-art verifiers on a set of large-scale benchmarks. The experiments show that our framework is both efficient and scalable even when applied to verify the robustness of large-scale neural networks.

AB - Robustness verification of ‘s is becoming increasingly crucial for their potential use in many safety-critical applications. Essentially, the problem of robustness verification can be encoded as a typical Mixed-Integer Linear Programming (MILP) problem, which can be solved via branch-and-bound strategies. However, these methods can only afford limited scalability and remain challenging for verifying large-scale neural networks. In this paper, we present a novel framework to speed up the solving of the MILP problems generated from the robustness verification of deep neural networks. It employs a semi-planet relaxation to abstract ReLU activation functions, via an RNN-based strategy for selecting the relaxed ReLU neurons to be tightened. We have developed a prototype tool L2T and conducted comparison experiments with state-of-the-art verifiers on a set of large-scale benchmarks. The experiments show that our framework is both efficient and scalable even when applied to verify the robustness of large-scale neural networks.

KW - Learning methods

KW - Neural networks

KW - Robustness verification

KW - Semi-planet relaxation

UR - https://www.scopus.com/pages/publications/85151048673

U2 - 10.1007/978-3-031-26319-4_34

DO - 10.1007/978-3-031-26319-4_34

M3 - 会议稿件

AN - SCOPUS:85151048673

SN - 9783031263187

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 571

EP - 586

BT - Computer Vision – ACCV 2022 - 16th Asian Conference on Computer Vision, 2022, Proceedings

A2 - Wang, Lei

A2 - Gall, Juergen

A2 - Chin, Tat-Jun

A2 - Sato, Imari

A2 - Chellappa, Rama

PB - Springer Science and Business Media Deutschland GmbH

T2 - 16th Asian Conference on Computer Vision, ACCV 2022

Y2 - 4 December 2022 through 8 December 2022

ER -

Xue H, Zeng X, Lin W, Yang Z , Peng C, Zeng Z. An RNN-Based Framework for the MILP Problem in Robustness Verification of Neural Networks. In Wang L, Gall J, Chin TJ, Sato I, Chellappa R, editors, Computer Vision – ACCV 2022 - 16th Asian Conference on Computer Vision, 2022, Proceedings. Springer Science and Business Media Deutschland GmbH. 2023. p. 571-586. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-031-26319-4_34