Evasion Attack against CNN-Adaboost Electricity Theft Detection in Smart Grid DOI

Santosh Nirmal,

Pramod Patil

Опубликована: Окт. 25, 2024

Язык: Английский

Adaptive Backdoor Attacks Against Dataset Distillation for Federated Learning DOI
Ze Chai, Zhipeng Gao,

Yijing Lin

и другие.

ICC 2022 - IEEE International Conference on Communications, Год журнала: 2024, Номер unknown, С. 4614 - 4619

Опубликована: Июнь 9, 2024

Язык: Английский

Процитировано

0
Risk of Text Backdoor Attacks Under Dataset Distillation DOI

Kejun Zhang,

Yutuo Song,

Shaofei Xu

и другие.

Lecture notes in computer science, Год журнала: 2024, Номер unknown, С. 127 - 144

Опубликована: Окт. 17, 2024

Язык: Английский

Процитировано

0
Secured energy data transaction for prosumers under diverse cyberattack scenarios DOI Creative Commons
Fariya Tabassum, Md. Rashidul Islam, M. Imran Azim

и другие.

Sustainable Energy Grids and Networks, Год журнала: 2024, Номер unknown, С. 101555 - 101555

Опубликована: Окт. 1, 2024

Язык: Английский

Процитировано

0
Data Poisoning Attack on Black-Box Neural Machine Translation to Truncate Translation DOI Creative Commons
Lingfang Li, Weijian Hu, M. X. Luo

и другие.

Entropy, Год журнала: 2024, Номер 26(12), С. 1081 - 1081

Опубликована: Дек. 11, 2024

Neural machine translation (NMT) systems have achieved outstanding performance and been widely deployed in the real world. However, undertranslation problem caused by distribution of high-translation-entropy words source sentences still exists, can be aggravated poisoning attacks. In this paper, we propose a new backdoor attack on NMT models small fraction parallel training data. Our increases entropy after injecting trigger, making them more easily discarded NMT. The final is part target translation, position injected trigger poison affects scope truncation. Moreover, also defense method, Backdoor Defense Sematic Representation Change (BDSRC), against our attack. Specifically, selected candidates based similarity between semantic representation sentence overall representation. Then, identified through computing deviation candidates. experiments show that strategy achieve nearly 100% success rate, functionality main tasks almost unaffected having degradation less than 1 BLEU. Nonetheless, method effectively identify triggers alleviate degradation.

Язык: Английский

Процитировано

0
Evasion Attack against CNN-Adaboost Electricity Theft Detection in Smart Grid DOI

Santosh Nirmal,

Pramod Patil

Опубликована: Окт. 25, 2024

Язык: Английский

Процитировано

0