Research map / Adversarial robustness and security / Backdoor and poisoning
One-shot Neural Backdoor Erasing via Adversarial Weight Masking
Shuwen Chai, Jinghui Chen
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 |
|---|---|---|---|
| replace neuron pruning with soft weight masking and recover adversarial triggers | backdoor removal on CIFAR-10 · CIFAR-10 one-shot, BadNets | ASR: 32.83 -> 8.98 | Improved |
| replace neuron pruning with soft weight masking and recover adversarial triggers | backdoor removal on CIFAR-10 · CIFAR-10 one-shot, BadNets | ACC: 60.35 -> 76.46 | Improved |
| add soft weight masking to ANP (ablation) | ANP · CIFAR-10 one-shot, BadNets | ACC: 53.26 -> 73.34 | Improved |
| add adversarial trigger recovery to ANP + SWM (ablation) | ANP + SWM · CIFAR-10 one-shot, BadNets | ACC: 73.34 -> 80.38 | Improved |
| add adversarial trigger recovery to ANP + SWM (ablation) | ANP + SWM · CIFAR-10 one-shot, BadNets | ASR: 10.16 -> 10.41 | No clear effect |
| add L1 mask regularization to ANP + SWM + ATR (ablation) | ANP + SWM + ATR · CIFAR-10 one-shot, BadNets | ASR: 10.41 -> 8.98 | Improved |
| remove clipping of the recovered perturbation (ablation) | AWM · CIFAR-10 one-shot, BadNets | ASR: 8.98 -> 16.28 | Hurt |
| remove clipping of the recovered perturbation (ablation) | AWM · CIFAR-10 one-shot, BadNets | ACC: 76.46 -> 66.03 | Hurt |
| remove L1 regularization on the mask (ablation) | AWM · CIFAR-10 one-shot, BadNets | ACC: 76.46 -> 66.32 | Hurt |
| remove perturbation clipping and mask L1 regularization (ablation) | AWM · CIFAR-10 one-shot, BadNets | ACC: 76.46 -> 65.32 | Hurt |
| replace the L1 perturbation constraint with an L2 constraint (ablation) | AWM · CIFAR-10 one-shot, BadNets | ASR: 8.98 -> 8.98 | No clear effect |
| use an L2 perturbation constraint and remove perturbation clipping (ablation) | AWM · CIFAR-10 one-shot, BadNets | ASR: 8.98 -> 14.38 | Hurt |
Closest papers by meaning
- Progressive Backdoor Erasing via Connecting Backdoor and Adversarial Attacks (CVPR 2023 · similarity 0.90)
- Injection, Attack and Erasure: Revocable Backdoor Attacks via Machine Unlearning (AAAI 2026 · similarity 0.88)
- From Toxic to Trustworthy: Using Self-Distillation and Semi-supervised Methods to Refine Neural Networks (AAAI 2024 · similarity 0.87)
- BAN: Detecting Backdoors Activated by Adversarial Neuron Noise (NeurIPS 2024 · similarity 0.87)
- Unveiling and Mitigating Backdoor Vulnerabilities based on Unlearning Weight Changes and Backdoor Activeness (NeurIPS 2024 · similarity 0.87)
- Adversarial-Inspired Backdoor Defense via Bridging Backdoor and Adversarial Attacks (AAAI 2025 · 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.
