中国科学院机构知识库网格
Chinese Academy of Sciences Institutional Repositories Grid
首页 机构 成果 学者
  
  1. CAS IR Grid
  2. 自动化研究所
  3. 中国科学院自动化研究所
  4. 多模态人工智能系统全国重点实验室
  5. 互联网大数据与安全信息学研究中心
ASCL: Adversarial supervised contrastive learning for defense against word substitution attacks

文献类型：期刊论文

作者Shi, Jiahui1,3; Li, Linjing1,2,3,4; Zeng, Daniel1,3
刊名NEUROCOMPUTING
出版日期2022-10-21
卷号510页码:59-68
ISSN号0925-2312
关键词Adversarial example Adversarial training Model robustness Contrastive learning Natural language processing
DOI10.1016/j.neucom.2022.09.032
通讯作者Li, Linjing(linjing.li@ia.ac.cn)
英文摘要Attacks with adversarial examples can tremendously worsen the performance of deep neural networks (DNNs). Hence, defending against such adversarial attacks is crucial for nearly all DNN-based applica-tions. Adversarial training is an effective and extensively adopted approach for increasing the robustness of DNNs in which benign examples and their adversarial counterparts are considered together in the training stage. However, this may result in a decrease in accuracy on benign examples because it does not account for the inter-class distance of benign examples. To overcome the aforementioned dilemma, we devise a novel defense approach named adversarial supervised contrastive learning (ASCL), which combines adversarial training with supervised contrastive learning to enhance the robustness of DNN-based models while maintaining their clean accuracy. We validate the effectiveness of the proposed ASCL approach in the scenario of defending against word substitution attacks by means of extensive experiments on benchmark tasks and datasets. The experimental results show that ASCL reduces the attack success rate to 20% while maintaining the accuracy for clean inputs within a 2% margin. (c) 2022 Elsevier B.V. All rights reserved.
资助项目National Key Research and Development Program of China[662020AAA0103405] ; National Natural Science Foundation of China[71621002] ; National Natural Science Foundation of China[62206282] ; Strategic Priority Research Program of Chinese Academy of Sciences[XDA27030100]
WOS研究方向Computer Science
语种英语
出版者ELSEVIER
WOS记录号WOS:000862258000006
资助机构National Key Research and Development Program of China ; National Natural Science Foundation of China ; Strategic Priority Research Program of Chinese Academy of Sciences
源URL[http://ir.ia.ac.cn/handle/173211/50433]  
专题自动化研究所_复杂系统管理与控制国家重点实验室_互联网大数据与安全信息学研究中心
通讯作者Li, Linjing
作者单位1.Chinese Acad Sci, Inst Automat, State Key Lab Management & Control Complex Syst, Beijing 100190, Peoples R China
2.Tianjin Zhongke Intelligent Recognit Co Ltd, Tianjin 300450, Peoples R China
3.Univ Chinese Acad Sci, Sch Artificial Intelligence, Beijing 101408, Peoples R China
4.Chinese Acad Sci, Inst Automat, Beijing, Peoples R China
推荐引用方式
GB/T 7714		 Shi, Jiahui,Li, Linjing,Zeng, Daniel. ASCL: Adversarial supervised contrastive learning for defense against word substitution attacks[J]. NEUROCOMPUTING,2022,510:59-68.
APA Shi, Jiahui,Li, Linjing,&Zeng, Daniel.(2022).ASCL: Adversarial supervised contrastive learning for defense against word substitution attacks.NEUROCOMPUTING,510,59-68.
MLA Shi, Jiahui,et al."ASCL: Adversarial supervised contrastive learning for defense against word substitution attacks".NEUROCOMPUTING 510(2022):59-68.

入库方式： OAI收割

来源：自动化研究所

浏览0
下载0
收藏0
其他版本

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。