Research map / Label-efficient and robust learning / Knowledge distillation
Undistillable: Making A Nasty Teacher That CANNOT teach students
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 self-undermining training to a normally trained teacher | ResNet-18 · CIFAR-10 standard KD | student accuracy: 87.75 (+1.11) -> 82.46 (-4.18) | Hurt |
| Add self-undermining training to a normally trained teacher | ResNeXt-29 · CIFAR-100 standard KD with MobileNetV2 | student accuracy: 72.43 (+3.31) -> 1.55 (-67.57) | Hurt |
| Add self-undermining training to a normally trained teacher | ResNet-18 · Tiny-ImageNet standard KD with MobileNetV2 | student accuracy: 55.99 (+4.27) -> 1.82 (-49.90) | Hurt |
| Add self-undermining training to a normally trained teacher | ResNet-34 · CIFAR-10 data-free KD with DAFL | student accuracy: 92.49 -> 86.15 (-6.34) | Hurt |
| Add self-undermining training to a normally trained teacher | ResNet-34 · CIFAR-100 data-free KD with DAFL | student accuracy: 71.06 -> 65.67 (-5.39) | Hurt |
| Use CNN instead of ResNet-18 as the adversarial network (ablation) | Self-undermining ResNet-18 teacher · CIFAR-10 standard KD to CNN | student accuracy: 82.46 (-4.18) -> 77.12 (-9.52) | Helped |
| Use ResNeXt-29 instead of ResNet-18 as the adversarial network (ablation) | Self-undermining ResNet-18 teacher · CIFAR-10 standard KD to CNN | student accuracy: 82.46 (-4.18) -> 82.75 (-3.89) | No clear effect |
| Use a larger student, ResNet-50, for reverse KD (ablation) | ResNet-18 normal teacher · CIFAR-10 | student accuracy: 94.98 -> 94.45 (-0.53) | Hurt |
| Use a larger student, ResNet-50, with a nasty teacher (ablation) | ResNet-18 teacher · CIFAR-10 | student accuracy: 94.98 -> 93.13 (-1.85) | Hurt |
| Increase the self-undermining weight ω (ablation) | Nasty teacher · CIFAR-10 and CIFAR-100 | student accuracy | Mixed |
| Vary distillation temperature τ_s from 1 to 20 (ablation) | Nasty teacher · Knowledge distillation | student accuracy | Mixed |
| Vary student KD balance factor α from 0.1 to 1.0 (ablation) | Student distilling from a nasty teacher · Knowledge distillation | student accuracy | Mixed |
Closest papers by meaning
- Can Students Beyond the Teacher? Distilling Knowledge from Teacher’s Bias (AAAI 2025 · similarity 0.91)
- Speculative Knowledge Distillation: Bridging the Teacher-Student Gap Through Interleaved Sampling (ICLR 2025 · similarity 0.89)
- All You Need in Knowledge Distillation Is a Tailored Coordinate System (AAAI 2025 · similarity 0.88)
- Teacher as a Lenient Expert: Teacher-Agnostic Data-Free Knowledge Distillation (AAAI 2024 · similarity 0.88)
- Taught Well Learned Ill: Towards Distillation-conditional Backdoor Attack (NeurIPS 2025 · similarity 0.88)
- Knowledge Distillation with Auxiliary Variable (ICML 2024 · 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.