A CMA-ES-Based Adversarial Attack on Black-Box Deep Neural Networks
Deep neural networks(DNNs) are widely used in AI-controlled Cyber-Physical Systems (CPS) to controll cars, robotics, water treatment plants and railways. However, DNNs have vulnerabilities to well-designed input samples that are called adversarial examples. Adversary attack is one of the important t...
Main Authors: | , , , , , |
---|---|
Format: | Article |
Language: | English |
Published: |
IEEE
2019-01-01
|
Series: | IEEE Access |
Subjects: | |
Online Access: | https://ieeexplore.ieee.org/document/8917642/ |
id |
doaj-a702e56d302e4867983d9670f72980ba |
---|---|
record_format |
Article |
spelling |
doaj-a702e56d302e4867983d9670f72980ba2021-03-30T00:50:35ZengIEEEIEEE Access2169-35362019-01-01717293817294710.1109/ACCESS.2019.29565538917642A CMA-ES-Based Adversarial Attack on Black-Box Deep Neural NetworksXiaohui Kuang0https://orcid.org/0000-0003-3816-402XHongyi Liu1https://orcid.org/0000-0002-3651-6960Ye Wang2https://orcid.org/0000-0001-7024-0067Qikun Zhang3https://orcid.org/0000-0003-2980-7695Quanxin Zhang4https://orcid.org/0000-0002-5094-7388Jun Zheng5https://orcid.org/0000-0002-1947-0921School of Computer Science, Beijing Institute of Technology, Beijing, ChinaSchool of Software and Microelectronics, Peking University, Beijing, ChinaDevelopment and Service Center for Science and Technology Talents, Ministry of Science Technology (MoST), Beijing, Exchange, ChinaSchool of Computer and Communication Engineering, Zhengzhou University of Light Industry, Zhengzhou, ChinaSchool of Computer Science, Beijing Institute of Technology, Beijing, ChinaSchool of Computer Science, Beijing Institute of Technology, Beijing, ChinaDeep neural networks(DNNs) are widely used in AI-controlled Cyber-Physical Systems (CPS) to controll cars, robotics, water treatment plants and railways. However, DNNs have vulnerabilities to well-designed input samples that are called adversarial examples. Adversary attack is one of the important techniques for detecting and improving the security of neural networks. Existing attacks, including state-of-the-art black-box attack have a lower success rate and make invalid queries that are not beneficial to obtain the direction of generating adversarial examples. For these reasons, this paper proposed a CMA-ES-based adversarial attack on black-box DNNs. Firstly, an efficient method to reduce the number of invalid queries is introduced. Secondly, a black-box attack of generating adversarial examples to fit a high-dimensional independent Gaussian distribution of the local solution space is proposed. Finally, a new CMA-based perturbation compression method is applied to make the process of reducing perturbation smoother. Experimental results on ImageNet classifiers show that the proposed attack has a higher success-rate than the state-of-the-art black-box attack but reduce the number of queries by 30% equally.https://ieeexplore.ieee.org/document/8917642/Deep neural networksadversarial exampleblack-box attackevolutionary strategy |
collection |
DOAJ |
language |
English |
format |
Article |
sources |
DOAJ |
author |
Xiaohui Kuang Hongyi Liu Ye Wang Qikun Zhang Quanxin Zhang Jun Zheng |
spellingShingle |
Xiaohui Kuang Hongyi Liu Ye Wang Qikun Zhang Quanxin Zhang Jun Zheng A CMA-ES-Based Adversarial Attack on Black-Box Deep Neural Networks IEEE Access Deep neural networks adversarial example black-box attack evolutionary strategy |
author_facet |
Xiaohui Kuang Hongyi Liu Ye Wang Qikun Zhang Quanxin Zhang Jun Zheng |
author_sort |
Xiaohui Kuang |
title |
A CMA-ES-Based Adversarial Attack on Black-Box Deep Neural Networks |
title_short |
A CMA-ES-Based Adversarial Attack on Black-Box Deep Neural Networks |
title_full |
A CMA-ES-Based Adversarial Attack on Black-Box Deep Neural Networks |
title_fullStr |
A CMA-ES-Based Adversarial Attack on Black-Box Deep Neural Networks |
title_full_unstemmed |
A CMA-ES-Based Adversarial Attack on Black-Box Deep Neural Networks |
title_sort |
cma-es-based adversarial attack on black-box deep neural networks |
publisher |
IEEE |
series |
IEEE Access |
issn |
2169-3536 |
publishDate |
2019-01-01 |
description |
Deep neural networks(DNNs) are widely used in AI-controlled Cyber-Physical Systems (CPS) to controll cars, robotics, water treatment plants and railways. However, DNNs have vulnerabilities to well-designed input samples that are called adversarial examples. Adversary attack is one of the important techniques for detecting and improving the security of neural networks. Existing attacks, including state-of-the-art black-box attack have a lower success rate and make invalid queries that are not beneficial to obtain the direction of generating adversarial examples. For these reasons, this paper proposed a CMA-ES-based adversarial attack on black-box DNNs. Firstly, an efficient method to reduce the number of invalid queries is introduced. Secondly, a black-box attack of generating adversarial examples to fit a high-dimensional independent Gaussian distribution of the local solution space is proposed. Finally, a new CMA-based perturbation compression method is applied to make the process of reducing perturbation smoother. Experimental results on ImageNet classifiers show that the proposed attack has a higher success-rate than the state-of-the-art black-box attack but reduce the number of queries by 30% equally. |
topic |
Deep neural networks adversarial example black-box attack evolutionary strategy |
url |
https://ieeexplore.ieee.org/document/8917642/ |
work_keys_str_mv |
AT xiaohuikuang acmaesbasedadversarialattackonblackboxdeepneuralnetworks AT hongyiliu acmaesbasedadversarialattackonblackboxdeepneuralnetworks AT yewang acmaesbasedadversarialattackonblackboxdeepneuralnetworks AT qikunzhang acmaesbasedadversarialattackonblackboxdeepneuralnetworks AT quanxinzhang acmaesbasedadversarialattackonblackboxdeepneuralnetworks AT junzheng acmaesbasedadversarialattackonblackboxdeepneuralnetworks AT xiaohuikuang cmaesbasedadversarialattackonblackboxdeepneuralnetworks AT hongyiliu cmaesbasedadversarialattackonblackboxdeepneuralnetworks AT yewang cmaesbasedadversarialattackonblackboxdeepneuralnetworks AT qikunzhang cmaesbasedadversarialattackonblackboxdeepneuralnetworks AT quanxinzhang cmaesbasedadversarialattackonblackboxdeepneuralnetworks AT junzheng cmaesbasedadversarialattackonblackboxdeepneuralnetworks |
_version_ |
1724187732854439936 |