Research map / Label-efficient and robust learning / Self-supervised and contrastive
Detecting Backdoors in Pre-Trained Encoders
Shiwei Feng, Guanhong Tao, Siyuan Cheng, Guangyu Shen, Xiangzhe Xu, Yingqi Liu, Kaiyuan Zhang, Shiqing Ma, Xiangyu Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use minimal-trigger inversion to detect backdoored encoders | DECREE · Encoders across GTSRB, SVHN, and STL-10 attack targets | detection accuracy: 99.1; 97.7; 99.1 | Improved |
| Use a shadow dataset instead of the pre-training dataset for trigger inversion | DECREE · CIFAR10-pretrained encoders, STL-10 shadow dataset | separation of clean and backdoored encoders by PL1-Norm | Improved |
| Add a similarity-loss term that encourages triggered inputs to be less similar | Original SSL backdoor attack · CIFAR10 encoder, SVHN target, downstream STL-10 | attack success rate: 99.9% to 69.9% | Hurt |
| Add a similarity-loss term that encourages triggered inputs to be less similar | Original SSL backdoor attack · CIFAR10 encoder, SVHN target | PL1-Norm: 0.14 | Mixed |
| Use Text-on-Pair attacks with natural-language target captions | DECREE · CLIP, SVHN attack target | detection accuracy: 87.5 | Hurt |
Closest papers by meaning
- DeDe: Detecting Backdoor Samples for SSL Encoders via Decoders (CVPR 2025 · similarity 0.89)
- Invisible Backdoor Attack against Self-supervised Learning (CVPR 2025 · similarity 0.89)
- Stealthy Backdoor Attack in Self-Supervised Learning Vision Encoders for Large Vision Language Models (CVPR 2025 · similarity 0.89)
- Not All Prompts Are Secure: A Switchable Backdoor Attack Against Pre-trained Vision Transfomers (CVPR 2024 · similarity 0.87)
- Temporal-Distributed Backdoor Attack against Video Based Action Recognition (AAAI 2024 · similarity 0.87)
- Clean Data Can Still Carry Backdoors: Support-Persistent Backdoors for Model Reuse (NeurIPS 2026 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
