Bypassing Backdoor Detection Algorithms in Deep Learning

被引:72
|
作者
Tan, Te Juin Lester [1 ]
Shokri, Reza [1 ]
机构
[1] Natl Univ Singapore NUS, Dept Comp Sci, Singapore, Singapore
关键词
D O I
10.1109/EuroSP48549.2020.00019
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 learning models are vulnerable to various adversarial manipulations of their training data, parameters, and input sample. In particular, an adversary can modify the training data and model parameters to embed backdoors into the model, so the model behaves according to the adversary's objective if the input contains the backdoor features, referred to as the backdoor trigger (e.g., a stamp on an image). The poisoned model's behavior on clean data, however, remains unchanged. Many detection algorithms are designed to detect backdoors on input samples or model parameters, through the statistical difference between the latent representations of adversarial and clean input samples in the poisoned model. In this paper, we design an adversarial backdoor embedding algorithm that can bypass the existing detection algorithms including the state-of-the-art techniques. We design an adaptive adversarial training algorithm that optimizes the original loss function of the model, and also maximizes the indistinguishability of the hidden representations of poisoned data and clean data. This work calls for designing adversary-aware defense mechanisms for backdoor detection.
引用
收藏
页码:175 / 183
页数:9
相关论文
共 50 条
  • [1] BIRD: Generalizable Backdoor Detection and Removal for Deep Reinforcement Learning
    Chen, Xuan
    Guo, Wenbo
    Tao, Guanhong
    Zhang, Xiangyu
    Song, Dawn
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 36 (NEURIPS 2023), 2023,
  • [2] NTD: Non-Transferability Enabled Deep Learning Backdoor Detection
    Li, Yinshan
    Ma, Hua
    Zhang, Zhi
    Gao, Yansong
    Abuadbba, Alsharif
    Xue, Minhui
    Fu, Anmin
    Zheng, Yifeng
    Al-Sarawi, Said F.
    Abbott, Derek
    IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, 2024, 19 : 104 - 119
  • [3] NTD: Non-Transferability Enabled Deep Learning Backdoor Detection
    School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing
    210094, China
    不详
    SA
    5005, Australia
    不详
    ACT
    2601, Australia
    不详
    WA
    6009, Australia
    不详
    518055, China
    IEEE Trans. Inf. Forensics Secur., 1600, (104-119):
  • [4] A Backdoor Embedding Method for Backdoor Detection in Deep Neural Networks
    Liu, Meirong
    Zheng, Hong
    Liu, Qin
    Xing, Xiaofei
    Dai, Yinglong
    UBIQUITOUS SECURITY, 2022, 1557 : 1 - 12
  • [5] Robust Backdoor Detection for Deep Learning via Topological Evolution Dynamics
    Mo, Xiaoxing
    Zhang, Yechao
    Zhang, Leo Yu
    Luo, Wei
    Sun, Nan
    Hu, Shengshan
    Gao, Shang
    Xiang, Yang
    45TH IEEE SYMPOSIUM ON SECURITY AND PRIVACY, SP 2024, 2024, : 2048 - 2066
  • [6] ASSET: Robust Backdoor Data Detection Across a Multiplicity of Deep Learning Paradigms
    Pan, Minzhou
    Zeng, Yi
    Lyu, Lingjuan
    Lin, Xue
    Jia, Ruoxi
    PROCEEDINGS OF THE 32ND USENIX SECURITY SYMPOSIUM, 2023, : 2725 - 2742
  • [7] Diabetes detection using deep learning algorithms
    Swapna, G.
    Vinayakumar, R.
    Soman, K. P.
    ICT EXPRESS, 2018, 4 (04): : 243 - 246
  • [8] Backdoor Attack and Defense on Deep Learning: A Survey
    Bai, Yang
    Xing, Gaojie
    Wu, Hongyan
    Rao, Zhihong
    Ma, Chuan
    Wang, Shiping
    Liu, Xiaolei
    Zhou, Yimin
    Tang, Jiajia
    Huang, Kaijun
    Kang, Jiale
    IEEE TRANSACTIONS ON COMPUTATIONAL SOCIAL SYSTEMS, 2025, 12 (01): : 404 - 434
  • [9] Backdoor Trigger Detection Using Adversarial Perturbation for Deep-learning Phishing Detector
    Nishiura, Koko
    Kimura, Tomotaka
    Cheng, Jun
    38TH INTERNATIONAL CONFERENCE ON INFORMATION NETWORKING, ICOIN 2024, 2024, : 506 - 508
  • [10] Research on Liveness Detection Algorithms Based on Deep Learning
    Fan, Ying
    Shi, Yilin
    Wang, Xianliang
    Yi, Haiyang
    PROCEEDINGS OF 2019 IEEE 10TH INTERNATIONAL CONFERENCE ON SOFTWARE ENGINEERING AND SERVICE SCIENCE (ICSESS 2019), 2019, : 366 - 371