Research map / Label-efficient and robust learning / Knowledge distillation
Comprehensive Knowledge Distillation with Causal Intervention
Xiang Deng, Zhongfei Zhang
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 sample representation distillation (ablation) | CID · CIFAR-100 WRN-40-2 -> WRN-16-2 | accuracy (%): +1.51% | Improved |
| add class representation distillation (ablation) | CID · CIFAR-100 WRN-40-2 -> WRN-16-2 | accuracy (%): +0.65% | Improved |
| use causal intervention P(Y|do(X)) instead of P(Y|X) (ablation) | CID · CIFAR-100 WRN-40-2 -> WRN-16-2 | accuracy (%): +0.70% | Improved |
| use normalized MSE (NM_MSE) instead of plain MSE for sample alignment (ablation) | CID · CIFAR-100 WRN-40-2 -> WRN-16-2 | accuracy (%) | Improved |
| distill last-layer features instead of other layers or all layers (ablation) | CID · CIFAR-100 WRN-40-2 -> WRN-16-2 | accuracy (%) | Improved |
| add CID to KD baseline | KD · CIFAR-10 WRN-16-4 -> WRN-16-1 | accuracy (%): 92.55±0.10 -> 92.95±0.10 | Improved |
| use CID instead of SRRL | SRRL · CIFAR-100 WRN-40-4 -> WRN-16-2 | accuracy (%): 75.96±0.21 -> 76.40±0.15 | Improved |
| use CID instead of CRD on ImageNet | CRD · ImageNet top-1 | accuracy (%): 71.4 -> 71.9 | Improved |
| use CID in cross-dataset transfer | CRD · CIFAR-100 to STL-10 | accuracy (%): 72.2 -> 72.5 | Improved |
| use CID with limited training data (20%) | SP · CIFAR-100, 20% data | accuracy (%): 60.35 -> 62.13 | Improved |
| add extra linear layer to competitor methods | competitor KD methods · CIFAR-100 | accuracy (%) | Hurt |
Closest papers by meaning
- Can Students Beyond the Teacher? Distilling Knowledge from Teacher’s Bias (AAAI 2025 · similarity 0.91)
- De-confounded Data-free Knowledge Distillation for Handling Distribution Shifts (CVPR 2024 · similarity 0.90)
- Knowledge Distillation with Auxiliary Variable (ICML 2024 · similarity 0.89)
- PLD: A Choice-Theoretic List-Wise Knowledge Distillation (NeurIPS 2025 · similarity 0.88)
- Debiased Distillation for Consistency Regularization (AAAI 2025 · similarity 0.88)
- Small Scale Data-Free Knowledge Distillation (CVPR 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.
