Research map / Adversarial robustness and security / Backdoor and poisoning
On the Trade-off between Adversarial and Backdoor Robustness
Cheng-Hsin Weng, Yan-Ting Lee, Shan-Hung (Brandon) Wu
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 |
|---|---|---|---|
| apply adversarial training | standard network · MNIST | backdoor attack success rate: 17.2% -> 67.2% | Hurt |
| apply adversarial training | standard network · CIFAR-10 | backdoor attack success rate: 64.1% -> 99.9% | Hurt |
| apply adversarial training | standard network · ImageNet | backdoor attack success rate: 3.9% -> 65.4% | Hurt |
| apply adversarial training | standard network · CIFAR-10 | clean accuracy: 90% -> 79.3% | Hurt |
| apply adversarial training | standard network · ImageNet | adversarial robustness: 0.1% -> 18.4% | Improved |
| add Lipschitz regularization or denoising layers without adversarial training | standard network · MNIST | adversarial robustness: 0% -> 0% | No clear effect |
| apply certified defense IBP | standard network · MNIST | backdoor attack success rate: 36.3% -> 92.4% | Hurt |
| shrink trigger to 1x1 on adversarially trained network | adversarially trained network · CIFAR-10 | backdoor success rate: 31.1% -> 69.8% | Mixed |
| use spectral signatures defense against clean-label attack on adversarially trained model | pre-training backdoor defense · CIFAR-10 | poison detection rate: 24.4% -> 10.6% | Hurt |
| apply neural cleansing to adversarially trained model vs standard model | post-training backdoor defense · CIFAR-10 complex watermark | backdoor success rate after defense: 39.3% -> 1.2% | Improved |
Closest papers by meaning
- Towards Deep Learning Models Resistant to Adversarial Attacks (ICLR 2018 · similarity 0.90)
- On the Vulnerability of Adversarially Trained Models Against Two-faced Attacks (ICLR 2024 · similarity 0.90)
- Increasing Confidence in Adversarial Robustness Evaluations (NeurIPS 2022 · similarity 0.90)
- Adversarial-Inspired Backdoor Defense via Bridging Backdoor and Adversarial Attacks (AAAI 2025 · similarity 0.89)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.88)
- Rethinking the Adversarial Robustness of Multi-Exit Neural Networks in an Attack-Defense Game (CVPR 2025 · similarity 0.88)
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.
