Boosting Black-Box Adversarial Attacks with Meta Learning

被引:0
|
作者
Fu, Junjie [1 ,2 ]
Sun, Jian [1 ,2 ]
Wang, Gang [1 ,2 ]
机构
[1] Beijing Inst Technol, Key Lab Intelligent Control & Decis Complex Syst, Beijing 10081, Peoples R China
[2] Beijing Inst Technol, Chongqing Innovat Ctr, Chongqing 401120, Peoples R China
基金
中国国家自然科学基金; 国家重点研发计划;
关键词
Deep neural networks; adversarial examples; black-box attack; meta adversarial perturbation; transferability;
D O I
暂无
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
Deep neural networks (DNNs) have achieved remarkable success in diverse fields. However, it has been demonstrated that DNNs are very vulnerable to adversarial examples even in black-box settings. A large number of black-box attack methods have been proposed to in the literature. However, those methods usually suffer from low success rates and large query counts, which cannot fully satisfy practical purposes. In this paper, we propose a hybrid attack method which trains meta adversarial perturbations (MAPs) on surrogate models and performs black-box attacks by estimating gradients of the models. Our method uses the meta adversarial perturbation as an initialization and subsequently trains any black-box attack method for several epochs. Furthermore, the MAPs enjoy favorable transferability and universality, in the sense that they can be employed to boost performance of other black-box adversarial attack methods. Extensive experiments demonstrate that our method can not only improve the attack success rates, but also reduces the number of queries compared to other methods.
引用
收藏
页码:7308 / 7313
页数:6
相关论文
共 50 条
  • [41] Adversarial Black-Box Attacks with Timing Side-Channel Leakage
    Nakai, Tsunato
    Suzuki, Daisuke
    Omatsu, Fumio
    Fujino, Takeshi
    IEICE TRANSACTIONS ON FUNDAMENTALS OF ELECTRONICS COMMUNICATIONS AND COMPUTER SCIENCES, 2021, E104A (01) : 143 - 151
  • [42] Black-box Adversarial Attacks on Commercial Speech Platforms with Minimal Information
    Zhene, Baolin
    Jiang, Peipei
    Wang, Qian
    Li, Qi
    Shen, Chao
    Wang, Cong
    Ge, Yunjie
    Teng, Qingyang
    Zhang, Shenyi
    CCS '21: PROCEEDINGS OF THE 2021 ACM SIGSAC CONFERENCE ON COMPUTER AND COMMUNICATIONS SECURITY, 2021, : 86 - 107
  • [43] Mitigating Black-Box Adversarial Attacks via Output Noise Perturbation
    Aithal, Manjushree B.
    Li, Xiaohua
    IEEE ACCESS, 2022, 10 : 12395 - 12411
  • [44] Simultaneously Optimizing Perturbations and Positions for Black-Box Adversarial Patch Attacks
    Wei, Xingxing
    Guo, Ying
    Yu, Jie
    Zhang, Bo
    IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 2023, 45 (07) : 9041 - 9054
  • [45] Improving Black-box Adversarial Attacks with a Transfer-based Prior
    Cheng, Shuyu
    Dong, Yinpeng
    Pang, Tianyu
    Su, Hang
    Zhu, Jun
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 32 (NIPS 2019), 2019, 32
  • [46] Guessing Smart: Biased Sampling for Efficient Black-Box Adversarial Attacks
    Brunner, Thomas
    Diehl, Frederik
    Le, Michael Truong
    Knoll, Alois
    2019 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV 2019), 2019, : 4957 - 4965
  • [47] Parsimonious Black-Box Adversarial Attacks via Efficient Combinatorial Optimization
    Moon, Seungyong
    An, Gaon
    Song, Hyun Oh
    INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 97, 2019, 97
  • [48] Black-box adversarial attacks against image quality assessment models
    Ran, Yu
    Zhang, Ao-Xiang
    Li, Mingjie
    Tang, Weixuan
    Wang, Yuan-Gen
    Expert Systems with Applications, 2025, 260
  • [49] White-box and Black-box Adversarial Attacks to Obstacle Avoidance in Mobile Robots
    Rano, Inaki
    Christensen, Anders Lyhne
    2023 EUROPEAN CONFERENCE ON MOBILE ROBOTS, ECMR, 2023, : 64 - 69
  • [50] AKD: Using Adversarial Knowledge Distillation to Achieve Black-box Attacks
    Lian, Xin
    Huang, Zhiqiu
    Wang, Chao
    2023 INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS, IJCNN, 2023,