USENIX Security2023Top-tier venue
Aliasing Backdoor Attacks on Pre-trained Models
Cheng'an Wei, Yeonjoon Lee, Kai Chen, Guozhu Meng, Peizhuo Lv
Abstract
Pre-trained deep learning models are widely used to train accurate models with limited data in a short time. To reduce computational costs, pre-trained neural networks often employ subsampling operations. However, recent studies have shown that these subsampling operations can cause aliasing issues, resulting in problems with generalization. Despite this knowledge, there is still a lack of research on the relationship between the aliasing of neural networks and security threats, such as adversarial attacks and backdoor attacks, which manipulate model predictions without the awareness of victims. In this paper, we propose the aliasing backdoor, a low-cost and data-free attack that threatens mainstream pre-trained models and transfers to all student models fine-tuned from them. The key idea is to create an aliasing error in the strided layers of the network and manipulate a benign input to a targeted intermediate representation. To evaluate the attack, we conduct experiments on image classification, face recognition, and speech recognition tasks. The results show that our approach can effectively attack mainstream models with a success rate of over 95%. Our research, based on the aliasing error caused by subsampling, reveals a fundamental security weakness of strided layers, which are widely used in modern neural network architectures. To the best of our knowledge, this is the first work to exploit the strided layers to launch backdoor attacks. 1 1 The source code of this work is publicly available at https://github. com/CassiniHuy/AliasingBackdoorAttack . 2 The word "strided" describes layers whose strides >1.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
5 top-tier papers have cited it since. Know what they found before your reviewers do.
Lune can draw on
the full paper · the top-tier papers that cite it · related work
Free to start. No credit card required.
Lineage
- Unveiling the Pitfalls of Data-Free Backdoor Detection Against Pre-Trained Models cites Neural Cleanse: Identifying and Mitigating Backdoor Attacks in Neural Networks
- Unveiling the Pitfalls of Data-Free Backdoor Detection Against Pre-Trained Models cites An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
- Unveiling the Pitfalls of Data-Free Backdoor Detection Against Pre-Trained Models cites ABS: Scanning Neural Networks for Back-doors by Artificial Brain Stimulation
- EmbedX: Embedding-Based Cross-Trigger Backdoor Attack Against Large Language Models cites Instruction Backdoor Attacks Against Customized LLMs
- ABS: Scanning Neural Networks for Back-doors by Artificial Brain Stimulation cites Neural Cleanse: Identifying and Mitigating Backdoor Attacks in Neural Networks
- ABS: Scanning Neural Networks for Back-doors by Artificial Brain Stimulation cites Trojaning Attack on Neural Networks
- The Ripple Effect: On Unforeseen Complications of Backdoor Attacks cites Instruction Backdoor Attacks Against Customized LLMs
Cited by top-tier papers5
- Instruction Backdoor Attacks Against Customized LLMsRui Zhang, Hongwei Li, Rui Wen, Wenbo Jiang et al.USENIX Security 2024 · 83 citations
- Demystifying RCE Vulnerabilities in LLM-Integrated AppsTong Liu, Zizhuang Deng, Guozhu Meng, Yuekang Li et al.CCS 2024 · 19 citations
- Unveiling the Pitfalls of Data-Free Backdoor Detection Against Pre-Trained ModelsQuan Zhao, Linkang Du, Yuntao Wang, Zhou Su et al.USENIX Security 2026
- EmbedX: Embedding-Based Cross-Trigger Backdoor Attack Against Large Language ModelsNan Yan, Yuqing Li, Xiong Wang, Jing Chen et al.USENIX Security 2025
- The Ripple Effect: On Unforeseen Complications of Backdoor AttacksRui Zhang, Yun Shen, Hongwei Li, Wenbo Jiang et al.ICML 2025
Builds on14
- An Image is Worth 16x16 Words: Transformers for Image Recognition at ScaleAlexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn et al.ICLR 2021 · 21,477 citations
- wav2vec 2.0: A Framework for Self-Supervised Learning of Speech RepresentationsAlexei Baevski, Yuhao Zhou, Abdelrahman Mohamed, Michael AuliNeurIPS 2020 · 9,451 citations
- Neural Cleanse: Identifying and Mitigating Backdoor Attacks in Neural NetworksBolun Wang, Yuanshun Yao, Shawn Shan, Huiying Li et al.S&P 2019 · 1,801 citations
- Trojaning Attack on Neural NetworksYingqi Liu, Shiqing Ma, Yousra Aafer, Wen-Chuan Lee et al.NDSS 2018 · 1,377 citations
- ABS: Scanning Neural Networks for Back-doors by Artificial Brain StimulationYingqi Liu, Wen-Chuan Lee, Guanhong Tao, Shiqing Ma et al.CCS 2019 · 531 citations
Related papers
- BadPre: Task-agnostic Backdoor Attacks to Pre-trained NLP Foundation ModelsKangjie Chen, Yuxian Meng, Xiaofei Sun, Shangwei Guo et al.ICLR 2022 · 133 citations
- Backdoor Pre-trained Models Can Transfer to AllLujia Shen, Shouling Ji, Xuhong Zhang, Jinfeng Li et al.CCS 2021 · 72 citations
- Purifier: Plug-and-play Backdoor Mitigation for Pre-trained Models Via Anomaly Activation SuppressionXiaoyu Zhang, Yulin Jin, Tao Wang, Jian Lou et al.ACM MM 2022 · 12 citations
- WaNet - Imperceptible Warping-based Backdoor AttackTuan Anh Nguyen, Anh Tuan TranICLR 2021
- Latent Backdoor Attacks on Deep Neural NetworksYuanshun Yao, Huiying Li, Haitao Zheng, Ben Y. ZhaoCCS 2019 · 465 citations
