Input-Relational Verification of Deep Neural Networks

被引:1
|
作者
Banerjee, Debangshu [1 ]
Xu, Changming [1 ]
Singh, Gagandeep [1 ,2 ]
机构
[1] Univ Illinois, Urbana, IL 61801 USA
[2] VMware Res, Palo Alto, CA USA
来源
PROCEEDINGS OF THE ACM ON PROGRAMMING LANGUAGES-PACMPL | 2024年 / 8卷 / PLDI期
关键词
Abstract Interpretation; Deep Learning; Relational Verification; ABSTRACT DOMAIN;
D O I
10.1145/3656377
中图分类号
TP31 [计算机软件];
学科分类号
081202 ; 0835 ;
摘要
We consider the verification of input-relational properties defined over deep neural networks (DNNs) such as robustness against universal adversarial perturbations, monotonicity, etc. Precise verification of these properties requires reasoning about multiple executions of the same DNN. We introduce a novel concept of difference tracking to compute the difference between the outputs of two executions of the same DNN at all layers. We design a new abstract domain, DiffPoly for efficient difference tracking that can scale large DNNs. DiffPoly is equipped with custom abstract transformers for common activation functions (ReLU, Tanh, Sigmoid, etc.) and affine layers and can create precise linear cross-execution constraints. We implement an input-relational verifier for DNNs called RaVeN which uses DiffPoly and linear program formulations to handle a wide range of input-relational properties. Our experimental results on challenging benchmarks show that by leveraging precise linear constraints defined over multiple executions of the DNN, RaVeN gains substantial precision over baselines on a wide range of datasets, networks, and input-relational properties.
引用
收藏
页数:27
相关论文
共 50 条
  • [41] IMPLICIT SALIENCY IN DEEP NEURAL NETWORKS
    Sun, Yutong
    Prabhushankar, Mohit
    AlRegib, Ghassan
    2020 IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING (ICIP), 2020, : 2915 - 2919
  • [42] MULTILINGUAL TRAINING OF DEEP NEURAL NETWORKS
    Ghoshal, Arnab
    Swietojanski, Pawel
    Renals, Steve
    2013 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP), 2013, : 7319 - 7323
  • [43] Hyperbolic Deep Neural Networks: A Survey
    Peng, Wei
    Varanka, Tuomas
    Mostafa, Abdelrahman
    Shi, Henglin
    Zhao, Guoying
    IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 2022, 44 (12) : 10023 - 10044
  • [44] Lossless Compression of Deep Neural Networks
    Serra, Thiago
    Kumar, Abhinav
    Ramalingam, Srikumar
    INTEGRATION OF CONSTRAINT PROGRAMMING, ARTIFICIAL INTELLIGENCE, AND OPERATIONS RESEARCH, CPAIOR 2020, 2020, 12296 : 417 - 430
  • [45] Continuously Constructive Deep Neural Networks
    Irsoy, Ozan
    Alpaydin, Ethem
    IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS, 2020, 31 (04) : 1124 - 1133
  • [46] Visual Genealogy of Deep Neural Networks
    Wang, Qianwen
    Yuan, Jun
    Chen, Shuxin
    Su, Hang
    Qu, Huamin
    Liu, Shixia
    IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS, 2020, 26 (11) : 3340 - 3352
  • [47] Fast learning in Deep Neural Networks
    Chandra, B.
    Sharma, Rajesh K.
    NEUROCOMPUTING, 2016, 171 : 1205 - 1215
  • [48] Transfer Entropy in Deep Neural Networks
    Andonie, R.
    Cataron, A.
    Moldovan, A.
    INTERNATIONAL JOURNAL OF COMPUTERS COMMUNICATIONS & CONTROL, 2025, 20 (01)
  • [49] A survey on the applications of Deep Neural Networks
    Latha, R. S.
    Sreekanth, G. R. R.
    Suganthe, R. C.
    Selvaraj, R. Esakki
    2021 INTERNATIONAL CONFERENCE ON COMPUTER COMMUNICATION AND INFORMATICS (ICCCI), 2021,
  • [50] Deep support vector neural networks
    Diaz-Vico, David
    Prada, Jesus
    Omari, Adil
    Dorronsoro, Jose
    INTEGRATED COMPUTER-AIDED ENGINEERING, 2020, 27 (04) : 389 - 402