Clean-label poisoning attack with perturbation causing dominant features

Zhang, Chen; Tang, Zhengqiang; Li, Kenli

doi:10.1016/j.ins.2023.03.124

Cited by 6 publications

(1 citation statement)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In [19], ATT_FLAV is introduced, which is a framework that enhances the robustness of federated learning-based autonomous driving models against poisoning attacks by using a bandit-based AttackRegion-UCB algorithm to dynamically choose the target attack label region in each round of training. In [20], scholars introduce a new type of data poisoning attack designed to preserve personal data privacy, that can also be used as a powerful clean-label backdoor attack. The attack operates by adding unnoticeable perturbations to clean data, to confuse DNNs into making incorrect classifications.…”

Section: Z Wang Et Al / An Overview Of Artificial Intelligence Securi...mentioning

confidence: 99%

An Overview of Artificial Intelligence Security Issues

Wang,

Dong,

Xiang

et al. 2024

Frontiers in Artificial Intelligence and Applications

View full text Add to dashboard Cite

Artificial intelligence (AI) technology promotes human civilization, while it also raises concerns, mainly related to security. AI plays a double-edged sword role in the network space closely related to human life. On the one hand, AI assists in network protection by intelligently detecting network intrusion, reducing missed alarm rates and false alarm rates, forecasting network threats, automatically searching for malicious code, and supporting network defense. On the other hand, AI can also assist in network attacks, such as attacking voice recognition systems and malicious software detection. AI algorithms and computational data pose many security threats, affecting the security of face recognition, speech detection, malware detection, automated driving, and other security applications based on AI technology.

show abstract

Section: Z Wang Et Al / An Overview Of Artificial Intelligence Securi...mentioning

confidence: 99%