logo
科言猫学术研究的AI总结
language
手机端手机端
  • Phone Feature微信小程序
简体中文简体中文
notice消息通知
博客博客
delete
arrow
返回

Rectifying Multi-Attack Adversarial Perturbations in Deep Neural Network based Image Classifier

delete2025-10-13
delete0
PRE
AI
Y
Yulong Wang
J
Jiaxuan Song
T
Tianxiang Li
X
Xin Yuan
H
Hong Li
N
Ni Wei
DOI：10.1145/3765757delete
delete原文链接
delete原文求助
delete分享
delete收藏
摘要

摘要

En 中文
图像分类的深度神经网络（DNNs）仍然容易受到对抗扰动的影响——这些微妙的输入操纵会导致灾难性的误分类。为解决这一问题，我们提出了对抗图像校正器（Adversarial Image Rectifier, AIR），这是一种受语言学启发的检测与缓解框架，通过在特征层面拦截和反转对抗扰动来增强DNN的鲁棒性。与现有防御方法不同，AIR无需预先了解攻击模式：它首先将DNN的层次化隐藏层特征图编码为具有语义结构的句子表示，然后通过这些句子中的“情感”异常来识别对抗输入——这是一种用于描述微妙对抗痕迹的语言学隐喻。关键在于，我们确定了对抗扰动主要传播的一个关键中间层，并训练了一个轻量级校正网络，在该层选择性消除对抗特征，同时保留良性语义。在Tiny-ImageNet、CIFAR-10、SVHN和MS COCO上的大量实验表明，AIR在防御已知攻击和未知攻击时，分别实现了高达95.02%和94.62%的校正率，显著超越了现有的防御技术。

期刊

A
ACM Transactions on Privacy and Security
IF:
2.8
论文数:
293
被引数:
770

机构

暂无机构信息
引用论文

引用论文

New Adversarial Image Detection Based on Sentiment Analysis
errIEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
IF8.9
err2024-10-01
err4
errOAAI
errWang, Yulong; Li, Tianxiang; Li, Shenghong; Yuan, Xin; Ni, Wei
err分享
err收藏
An Efficient Dual-Hierarchy t-SNE Minimization
errIEEE Transactions on Visualization and Computer Graphics
IF0
err2022-01-01
err0
errOAAI
errMark van de Ruit; Markus Billeter; Elmar Eisemann
err分享
err收藏
Adversarial Learning Targeting Deep Neural Network Classification: A Comprehensive Review of Defenses Against Attacks
errPROCEEDINGS OF THE IEEE
IF25.9
err2020-03-01
err128
errOAAI
errMiller, David J.; Xiang, Zhen; Kesidis, George
err分享
err收藏
Pixle: a fast and effective black-box attack based on rearranging pixels
err2022 International Joint Conference on Neural Networks (IJCNN)
IF0
err2022-07-18
err0
errOAAI
errJary Pomponi; Simone Scardapane; Aurelio Uncini
err分享
err收藏
Dispersed Pixel Perturbation-Based Imperceptible Backdoor Trigger for Image Classifier Models
errIEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY
IF8
err2022-01-01
err10
errOAAI
errWang, Yulong; Zhao, Minghui; Li, Shenghong; Yuan, Xin; Ni, Wei
err分享
err收藏
A Generic Improvement to Deep Residual Networks Based on Gradient Flow
errIEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
IF8.9
err2020-01-01
err8
PREAI
errSanthanam, Venkataraman; Davis, Larry S.
err分享
err收藏
Towards Transferable Adversarial Attack Against Deep Face Recognition
errIEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY
IF8
err2021-01-01
err102
errOAAI
errZhong, Yaoyao; Deng, Weihong
err分享
err收藏
ADDITION: Detecting Adversarial Examples With Image-Dependent Noise Reduction
errIEEE TRANSACTIONS ON DEPENDABLE AND SECURE COMPUTING
IF7.5
err2024-05-01
err3
PREAI
errWang, Yuchen; Li, Xiaoguang; Yang, Li; Ma, Jianfeng; Li, Hui
err分享
err收藏
One Pixel Attack for Fooling Deep Neural Networks愚弄深度神经网络的一个像素攻击
errIEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION
IF12
err2019-10-01
err1.3K
errOAAI
errSu, Jiawei; Vargas, Danilo Vasconcellos; Sakurai, Kouichi
err分享
err收藏
Adversarial example detection based on saliency map features
errAPPLIED INTELLIGENCE
IF3.5
err2021-09-06
err15
PREAI
errWang, Shen; Gong, Yuxin
err分享
err收藏
没有更多内容