Research map / Adversarial robustness and security / Backdoor and poisoning
Trap and Replace: Defending Backdoor Attacks by Trapping Them into an Easy-to-Replace Subnetwork
Haotao Wang, Junyuan Hong, Aston Zhang, Jiayu Zhou, Zhangyang Wang
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 |
|---|---|---|---|
| remove auxiliary image reconstruction task in stage 1 (ablation) | Trap and Replace · CIFAR10 l2-Invisible | ASR (%): 0.72 -> 99.99 | Hurt |
| remove auxiliary image reconstruction task in stage 1 (ablation) | Trap and Replace · CIFAR10 Trojan-WM | ASR (%): 12.01 -> 51.87 | Hurt |
| do not replace the classification head (ablation) | Trap and Replace · CIFAR10 l2-Invisible | ASR (%): 0.72 -> 99.96 | Hurt |
| do not replace the classification head (ablation) | Trap and Replace · CIFAR10 SIG | ASR (%): 0.01 -> 97.24 | Hurt |
| stem has 13 layers, head only 1 FC layer (vs 11/3) (ablation) | Trap and Replace · CIFAR10 Blend | ASR (%): 10.59 -> 97.41 | Hurt |
| make head larger: 9 stem layers, 5 head layers (vs 11/3) (ablation) | Trap and Replace · CIFAR10 l2-Invisible | clean accuracy (%): 84.01 -> 73.50 | Hurt |
| make head larger: 9 stem layers, 5 head layers (vs 11/3) (ablation) | Trap and Replace · CIFAR10 Smooth | ASR (%): 4.23 -> 4.17 | No clear effect |
| raise reconstruction weight lambda1 from 0 to 10 (ablation) | Trap and Replace lambda1 · CIFAR10 l2-Invisible | ASR (%): 99.99 -> 0.74 | Improved |
| raise lambda1 from 10 to 20 (ablation) | Trap and Replace · CIFAR10 l2-Invisible | clean accuracy (%): 84.01 -> 81.37 | Hurt |
| set total variation weight lambda2 to 0.1 instead of 0 (ablation) | Trap and Replace · CIFAR10 l2-Invisible | ASR (%): 3.32 -> 0.74 | Improved |
| reduce clean holdout set from 2500 to 250 samples | Trap and Replace · CIFAR10 l2-Invisible | clean accuracy (%): 84.01 -> 72.00 | Hurt |
| reduce clean holdout set from 2500 to 1250 samples | Trap and Replace · CIFAR10 l2-Invisible | clean accuracy (%): 84.01 -> 83.31 | No clear effect |
Closest papers by meaning
- BaDExpert: Extracting Backdoor Functionality for Accurate Backdoor Input Detection (ICLR 2024 · similarity 0.89)
- BackdoorDM: A Comprehensive Benchmark for Backdoor Learning on Diffusion Model (NeurIPS 2025 · similarity 0.89)
- REFINE: Inversion-Free Backdoor Defense via Model Reprogramming (ICLR 2025 · similarity 0.89)
- Unveiling and Mitigating Backdoor Vulnerabilities based on Unlearning Weight Changes and Backdoor Activeness (NeurIPS 2024 · similarity 0.89)
- BAN: Detecting Backdoors Activated by Adversarial Neuron Noise (NeurIPS 2024 · similarity 0.89)
- Fusing Pruned and Backdoored Models: Optimal Transport-based Data-free Backdoor Mitigation (AAAI 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.
