Research map / Adversarial robustness and security / Backdoor and poisoning
BeDKD: Backdoor Defense Based on Directional Mapping Module and Adversarial Knowledge Distillation
Zhengxian Wu, Juan Wen, Wanli Peng, Yinghan Zhou, Changtong Dou, Yiming Xue
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 |
|---|---|---|---|
| add DMM-identified poisoned data to fine-tuning (FT+DMM) (ablation) | FT · SST2 (SynBkd) | attack success rate: 66.22 -> 22.97 | Improved |
| add DMM-identified poisoned data to KD (KD+DMM) (ablation) | KD · SST2 (SynBkd) | attack success rate: 94.60 -> 40.54 | Improved |
| use adversarial KD (trust+punish) instead of KD with DMM (ablation) | KD+DMM · SST2 (SynBkd) | attack success rate: 40.54 -> 2.48 | Improved |
| add Lsoft to DMM distillation loss (ablation) | DMM · SST2 (AddSent) | FAR: 76.13 -> 0.00 | Improved |
| add Lsoft to DMM distillation loss (ablation) | DMM · SST2 (SynBkd) | FAR: 66.67 -> 42.12 | Improved |
| increase clean samples per class from 320 to 640 (ablation) | BeDKD · SST2 (StyBkd) | attack success rate: 4.86 -> 10.09 | Hurt |
| increase clean samples per class from 80 to 320 (ablation) | BeDKD · SST2 (StyBkd) | clean accuracy: 87.84 -> 90.59 | Improved |
| BeDKD vs W2SDefense | BeDKD · OLID | attack success rate: 17.61 -> 1.27 | Improved |
| BeDKD vs W2SDefense | BeDKD · SST2 | attack success rate: 27.25 -> 1.79 | Improved |
| BeDKD vs W2SDefense | BeDKD · AGnews | attack success rate: 3.34 -> 0.81 | Improved |
| BeDKD vs no defense on clean accuracy | BeDKD · SST2 | clean accuracy: 90.27 -> 90.60 | No clear effect |
| lower poisoning rate from 10% to 5% on OLID | BeDKD · OLID | attack success rate: 1.35% -> 1.28% (average) | No clear effect |
Closest papers by meaning
- Adversarial-Inspired Backdoor Defense via Bridging Backdoor and Adversarial Attacks (AAAI 2025 · similarity 0.89)
- Mitigating Backdoor Attack by Injecting Proactive Defensive Backdoor (NeurIPS 2024 · similarity 0.88)
- Inspecting Prediction Confidence for Detecting Black-Box Backdoor Attacks (AAAI 2024 · similarity 0.88)
- Rethinking Backdoor Attacks on Dataset Distillation: A Kernel Method Perspective (ICLR 2024 · similarity 0.88)
- From Toxic to Trustworthy: Using Self-Distillation and Semi-supervised Methods to Refine Neural Networks (AAAI 2024 · similarity 0.88)
- Protecting against simultaneous data poisoning attacks (ICLR 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.
