Research map / Adversarial robustness and security / Backdoor and poisoning
Transferable Backdoor Attacks for Code Models via Sharpness-Aware Adversarial Perturbation
Shuyu Chang, Haiping Huang, Yanjun Zhang, Yujin Huang, Fu Xiao, Leo Yu Zhang
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 STAB instead of AFRAIDOOR | backdoor attack · MNP victim Py150 (PLBART) | ASR (%): 68.48 -> 76.89 | Improved |
| use STAB instead of AFRAIDOOR | backdoor attack · CS victim PyT (PLBART) | ASR (%): 78.37 -> 83.67 | Improved |
| use STAB instead of AFRAIDOOR | backdoor attack · MNP victim CSN (CodeT5) | ASR (%): 90.13 -> 95.13 | Improved |
| use STAB instead of AFRAIDOOR | backdoor attack · MNP Py150, KillBadCode defense (CodeT5) | detection recall (%): 27.20 -> 23.02 | Improved |
| add SAM to surrogate training (ablation) | STAB · MNP victim Py150 (PLBART) | ASR (%): 72.21 -> 76.89 | Improved |
| add SAM to surrogate training (ablation) | STAB · MNP victim Py150 (PLBART) | ASR-D (%): 63.92 -> 70.17 | Improved |
| add SAM to surrogate training (ablation) | STAB · MNP victim CSN (PLBART) | ASR (%): 91.12 -> 94.55 | Improved |
| add Gumbel-Softmax optimization instead of greedy search (ablation) | STAB · MNP victim Py150 (PLBART) | ASR-D (%): 67.31 -> 70.17 | Improved |
| add Gumbel-Softmax optimization instead of greedy search (ablation) | STAB · MNP victim PyT (PLBART) | ASR-D (%): 71.42 -> 72.01 | Improved |
| add Gumbel-Softmax optimization instead of greedy search (ablation) | STAB · MNP victims (PLBART) | ASR (%): 74.58 -> 76.89 (Py150); 93.85 -> 94.55 (CSN) | Mixed |
| use static Fixed or Grammar triggers under KillBadCode | static attacks · MNP Py150, KillBadCode defense (CodeT5) | ASR-D (%): 0.0 under defense | Hurt |
Closest papers by meaning
- Demystifying Poisoning Backdoor Attacks from a Statistical Perspective (ICLR 2024 · similarity 0.88)
- Understanding and Enhancing the Transferability of Jailbreaking Attacks (ICLR 2025 · similarity 0.87)
- REFINE: Inversion-Free Backdoor Defense via Model Reprogramming (ICLR 2025 · similarity 0.87)
- Universal Backdoor Attacks (ICLR 2024 · similarity 0.87)
- Revisiting Backdoor Attacks against Large Vision-Language Models from Domain Shift (CVPR 2025 · similarity 0.87)
- Unveiling and Mitigating Backdoor Vulnerabilities based on Unlearning Weight Changes and Backdoor Activeness (NeurIPS 2024 · 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.
