Research map / Label-efficient and robust learning / Knowledge distillation
Knowledge Condensation Distillation
Chenxin Li, Mingbao Lin, Zhiyuan Ding, Nie Lin, Yihong Zhuang, Yue Huang, Xinghao Ding, Liujuan Cao
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 OGVE ranking instead of random selection of knowledge points (ablation) | KCD · CIFAR100 WRN-40-2 -> WRN-16-2 | accuracy: 74.54 -> 75.48 | Improved |
| add online value recording (ablation) | KCD OGVE · CIFAR100 WRN-40-2 -> WRN-16-2 | accuracy: 75.01 -> 75.48 | Improved |
| add cost-aware ranking (ablation) | KCD OGVE · CIFAR100 WRN-40-2 -> WRN-16-2 | accuracy: 75.27 -> 75.48 | Improved |
| add VAKS on top of OGVE (ablation) | KCD · CIFAR100 WRN-40-2 -> WRN-16-2 | accuracy: 75.48 -> 75.70 | Improved |
| use adaptive perturbation ratio instead of fixed eps_m (ablation) | VAKS · CIFAR100 WRN-40-2 -> WRN-16-2 | accuracy: 75.57 -> 75.70 | Improved |
| add knowledge augmentation on the boundary set (ablation) | KCD · CIFAR100 WRN-40-2 -> WRN-16-2 | accuracy: 75.45 -> 75.70 | Improved |
| augment low-value K1L rather than high-value K1H (ablation) | VAKS · CIFAR100 WRN-40-2 -> WRN-16-2 | accuracy: 75.52 -> 75.70 | Improved |
| condense to rho=0.7 rather than 0.3 (ablation) | KCD · CIFAR100 WRN-40-2 -> WRN-16-2 | accuracy: 74.07 -> 75.70 | Improved |
| condense to rho=0.7 rather than 0.9 (ablation) | KCD · CIFAR100 WRN-40-2 -> WRN-16-2 | accuracy: 75.01 -> 75.70 | Improved |
| set alpha=0.03 rather than 0.01 (ablation) | KCD · CIFAR100 VGG13 -> MobileNetV2 | accuracy: 66.40 -> 68.61 | Improved |
| add knowledge condensation to KD | KD · CIFAR100 WRN-40-2 -> WRN-16-2 | accuracy: 74.92 -> 75.70 | Improved |
| add knowledge condensation to KD | KD · ImageNet ResNet34 -> ResNet18 | top-1 error: 29.34 -> 28.61 | Improved |
Closest papers by meaning
- A Good Teacher Adapts Their Knowledge for Distillation (ICCV 2025 · similarity 0.92)
- Can Students Beyond the Teacher? Distilling Knowledge from Teacher’s Bias (AAAI 2025 · similarity 0.92)
- Knowledge Distillation with Auxiliary Variable (ICML 2024 · similarity 0.92)
- Adaptive Dual Guidance Knowledge Distillation (AAAI 2025 · similarity 0.91)
- Toward Student-oriented Teacher Network Training for Knowledge Distillation (ICLR 2024 · similarity 0.91)
- Speculative Knowledge Distillation: Bridging the Teacher-Student Gap Through Interleaved Sampling (ICLR 2025 · similarity 0.90)
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.
