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
  • [1] DeepDyve: Dynamic Verification for Deep Neural Networks
    Li, Yu
    Li, Min
    Luo, Bo
    Tian, Ye
    Xu, Qiang
    CCS '20: PROCEEDINGS OF THE 2020 ACM SIGSAC CONFERENCE ON COMPUTER AND COMMUNICATIONS SECURITY, 2020, : 101 - 112
  • [2] ASVtorch toolkit: Speaker verification with deep neural networks
    Lee, Kong Aik
    Vestman, Ville
    Kinnunen, Tomi
    SOFTWAREX, 2021, 14
  • [3] An MILP Encoding for Efficient Verification of Quantized Deep Neural Networks
    Mistry, Samvid
    Saha, Indranil
    Biswas, Swarnendu
    IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS, 2022, 41 (11) : 4445 - 4456
  • [4] Acorns: A Framework for Accelerating Deep Neural Networks with Input Sparsity
    Dong, Xiao
    Liu, Lei
    Zhao, Peng
    Li, Guangli
    Li, Jiansong
    Wang, Xueying
    Feng, Xiaobing
    2019 28TH INTERNATIONAL CONFERENCE ON PARALLEL ARCHITECTURES AND COMPILATION TECHNIQUES (PACT 2019), 2019, : 178 - 191
  • [5] POSTER: Exploiting the Input Sparsity to Accelerate Deep Neural Networks
    Dong, Xiao
    Liu, Lei
    Li, Guangli
    Li, Jiansong
    Zhao, Peng
    Wang, Xueying
    Feng, Xiaobing
    PROCEEDINGS OF THE 24TH SYMPOSIUM ON PRINCIPLES AND PRACTICE OF PARALLEL PROGRAMMING (PPOPP '19), 2019, : 401 - 402
  • [6] Automatic Chinese Handwriting Verification Algorithm Using Deep Neural Networks
    Lee, Chi-Chang
    Ding, Jian-Jiun
    2019 INTERNATIONAL SYMPOSIUM ON INTELLIGENT SIGNAL PROCESSING AND COMMUNICATION SYSTEMS (ISPACS), 2019,
  • [7] Enhancing Robustness Verification for Deep Neural Networks via Symbolic Propagation
    Yang, Pengfei
    Li, Jianlin
    Liu, Jiangchao
    Huang, Cheng-Chao
    Li, Renjue
    Chen, Liqian
    Huang, Xiaowei
    Zhang, Lijun
    FORMAL ASPECTS OF COMPUTING, 2021, 33 (03) : 407 - 435
  • [8] Testing and Verification of the Deep Neural Networks Against Sparse Pixel Defects
    Szczepankiewicz, Michal
    Radlak, Krystian
    Szczepankiewicz, Karolina
    Popowicz, Adam
    Zawistowski, Pawel
    COMPUTER SAFETY, RELIABILITY, AND SECURITY, SAFECOMP 2022 WORKSHOPS, 2022, 13415 : 71 - 82
  • [9] Distance-Aware Test Input Selection for Deep Neural Networks
    Li, Zhong
    Xu, Zhengfeng
    Ji, Ruihua
    Pan, Minxue
    Zhang, Tian
    Wang, Linzhang
    Li, Xuandong
    PROCEEDINGS OF THE 33RD ACM SIGSOFT INTERNATIONAL SYMPOSIUM ON SOFTWARE TESTING AND ANALYSIS, ISSTA 2024, 2024, : 248 - 260
  • [10] Driver Identification and Verification From Smartphone Accelerometers Using Deep Neural Networks
    Hernandez Sanchez, Sara
    Fernandez Pozo, Ruben
    Hernandez Gomez, Luis Alfonso
    IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, 2022, 23 (01) : 97 - 109