Institutional Repository of Chinese Acad Sci, Inst Automat, Natl Lab Pattern Recognit, Beijing 100190, Peoples R China
| Adversarial training with distribution normalization and margin balance | |
Cheng, Zhen1,2; Zhu, Fei1,2 ; Zhang, Xu-Yao1,2; Liu, Cheng-Lin1,2
| |
| Source Publication | PATTERN RECOGNITION
 |
| ISSN | 0031-3203 |
| 2023-04-01 | |
| Volume | 136Pages:11 |
| Corresponding Author | Zhang, Xu-Yao(xyz@nlpr.ia.ac.cn) |
| Abstract | Adversarial training is the most effective method to improve adversarial robustness. However, it does not explicitly regularize the feature space during training. Adversarial attacks usually move a sample it-eratively along the direction which causes the steepest ascent of classification loss by crossing decision boundary. To alleviate this problem, we propose to regularize the distributions of different classes to increase the difficulty of finding an attacking direction. Specifically, we propose two strategies named Distribution Normalization (DN) and Margin Balance (MB) for adversarial training. The purpose of DN is to normalize the features of each class to have identical variance in every direction, in order to elimi-nate easy-to-attack intra-class directions. The purpose of MB is to balance the margins between different classes, making it harder to find confusing class directions (i.e., those with smaller margins) to attack. When integrated with adversarial training, our method can significantly improve adversarial robustness. Extensive experiments under white-box, black-box, and adaptive attacks demonstrate the effectiveness of our method over other state-of-the-art methods.(c) 2022 Elsevier Ltd. All rights reserved. |
| Keyword | Adversarial robustness Adversarial training Distribution normalization Margin balance |
| DOI | 10.1016/j.patcog.2022.109182 |
| Indexed By | SCI |
| Language | 英语 |
| Funding Project | National Key Research and Development Program[2018AAA0100400] ; National Natural Science Foundation of China (NSFC)[U20A20223] ; National Natural Science Foundation of China (NSFC)[62222609] ; National Natural Science Foundation of China (NSFC)[62076236] ; National Natural Science Foundation of China (NSFC)[61721004] ; Key Research Program of Frontier Sciences of Chinese Academy of Sciences[ZDBS-LY-7004] ; Youth Innovation Promotion Association of Chinese Academy of Sciences[2019141] |
| Funding Organization | National Key Research and Development Program ; National Natural Science Foundation of China (NSFC) ; Key Research Program of Frontier Sciences of Chinese Academy of Sciences ; Youth Innovation Promotion Association of Chinese Academy of Sciences |
| WOS Research Area | Computer Science ; Engineering |
| WOS Subject | Computer Science, Artificial Intelligence ; Engineering, Electrical & Electronic |
| WOS ID | WOS:000891819300004 |
| Publisher | ELSEVIER SCI LTD |
| Citation statistics | |
| Document Type | 期刊论文 |
| Identifier | http://ir.ia.ac.cn/handle/173211/50828 |
| Collection | 模式识别国家重点实验室_模式分析与学习 |
| Corresponding Author | Zhang, Xu-Yao |
| Affiliation | 1.Chinese Acad Sci, Natl Lab Pattern Recognit NLPR, Inst Automat, Beijing 100190, Peoples R China 2.Univ Chinese Acad Sci UCAS, Sch Artificial Intelligence, Beijing 100049, Peoples R China |
| First Author Affilication | Chinese Acad Sci, Inst Automat, Natl Lab Pattern Recognit, Beijing 100190, Peoples R China |
| Corresponding Author Affilication | Chinese Acad Sci, Inst Automat, Natl Lab Pattern Recognit, Beijing 100190, Peoples R China |
| Recommended Citation GB/T 7714 | Cheng, Zhen,Zhu, Fei,Zhang, Xu-Yao,et al. Adversarial training with distribution normalization and margin balance[J]. PATTERN RECOGNITION,2023,136:11. |
| APA | Cheng, Zhen,Zhu, Fei,Zhang, Xu-Yao,&Liu, Cheng-Lin.(2023).Adversarial training with distribution normalization and margin balance.PATTERN RECOGNITION,136,11. |
| MLA | Cheng, Zhen,et al."Adversarial training with distribution normalization and margin balance".PATTERN RECOGNITION 136(2023):11. |
| Files in This Item: | There are no files associated with this item. | |||||
Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.
Edit Comment