Research map / Label-efficient and robust learning / Knowledge distillation
Revisiting Data-Free Knowledge Distillation with Poisoned Teachers
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 ABD to ZSKT | ZSKT with WRN16-2 teacher · CIFAR-10, BadNets (grid) | ASR: 96.9 -> 0.7 | Helped |
| Add ABD to ZSKT | ZSKT with WRN16-2 teacher · CIFAR-10, Trojan WM | ASR: 93.9 -> 22.5 | Helped |
| Add ABD to ZSKT | ZSKT with WRN16-2 teacher · CIFAR-10, Trojan 3x3 | ASR: 96.8 -> 33.3 | Helped |
| Add ABD to ZSKT | ZSKT with WRN16-2 teacher · CIFAR-10, Sig | ASR: 0.0 -> 0. (s) | No clear effect |
| Add ABD to ZSKT | ZSKT with WRN16-2 teacher · CIFAR-10, l2_inv | ASR: 0.3 -> 1.9 (s) | Hurt |
| Add ABD to ZSKT | ZSKT with WRN16-2 teacher · GTSR-B, BadNets (grid) | ASR: 99.5 -> 13.0 | Helped |
| Add ABD to CMI | CMI with WRN16-2 teacher · CIFAR-10, Trojan WM | ASR: 99.0 -> 8.0 | Helped |
| Add ABD to OOD distillation | OOD with WRN16-2 teacher · CIFAR-10, Trojan WM | ASR: 100 -> 21.8 | Helped |
| Add Shuffling Vaccine without Self-Retrospection (ablation) | ZSKT · CIFAR-10, BadNets (grid) | ASR: 87.8 -> 0.3 | Helped |
| Add Self-Retrospection without Shuffling Vaccine (ablation) | ZSKT · CIFAR-10, BadNets (grid) | ASR: 87.8 -> 76.2 | Helped |
| Add Self-Retrospection after Shuffling Vaccine (ablation) | ZSKT with Shuffling Vaccine · CIFAR-10, Trojan WM | ASR: 57.0 -> 22.5 | Helped |
| Use Shuffling Vaccine without Self-Retrospection (ablation) | ZSKT · CIFAR-10, Trojan WM | ASR: 93.9 -> 57.0 | Helped |
Closest papers by meaning
- When Data-Free Knowledge Distillation Meets Non-Transferable Teacher: Escaping Out-of-Distribution Trap is All You Need (ICML 2025 · similarity 0.90)
- Taught Well Learned Ill: Towards Distillation-conditional Backdoor Attack (NeurIPS 2025 · similarity 0.89)
- Efficient Verified Unlearning For Distillation (NeurIPS 2025 · similarity 0.87)
- Rethinking Backdoor Attacks on Dataset Distillation: A Kernel Method Perspective (ICLR 2024 · similarity 0.87)
- AuG-KD: Anchor-Based Mixup Generation for Out-of-Domain Knowledge Distillation (ICLR 2024 · similarity 0.86)
- Teacher as a Lenient Expert: Teacher-Agnostic Data-Free Knowledge Distillation (AAAI 2024 · similarity 0.86)
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.