Research map / Label-efficient and robust learning / Knowledge distillation
Knowledge Distillation as Semiparametric Inference
Tri Dao, Govinda M Kamath, Vasilis Syrgkanis, Lester Mackey
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 |
|---|---|---|---|
| Apply cross-fitting instead of vanilla KD | Random forest student and teacher · FICO | AUC: up to 4 AUC percentage points | Improved |
| Apply loss correction instead of vanilla KD | Random forest student and teacher · Adult | student performance | Improved |
| Apply cross-fitting instead of vanilla KD | ResNet-8 student with ResNet teachers · CIFAR-10 | accuracy: up to 1.5% | Improved |
| Apply cross-fitting instead of vanilla KD | ResNet-8 student with ResNet teachers · CIFAR-10 | cross-entropy test loss | Improved |
| Add loss correction on top of cross-fitting | ResNet-8 student with ResNet teachers · CIFAR-10 | accuracy | Improved |
| Add loss correction on top of cross-fitting | ResNet-8 student with ResNet teachers · CIFAR-10 | cross-entropy test loss | Improved |
| Vary loss-correction strength from vanilla toward orthogonal correction (ablation) | Cross-fit KD with ResNet-8 student and ResNet-20 teacher · CIFAR-10 | test accuracy | Mixed |
| Use a very large loss-correction strength, approaching orthogonal correction (ablation) | Cross-fit KD with ResNet-8 student and ResNet-20 teacher · CIFAR-10 | test accuracy | Hurt |
| Use a very large loss-correction strength, approaching orthogonal correction (ablation) | Cross-fit KD with ResNet-8 student and ResNet-20 teacher · CIFAR-10 | training stability | Hurt |
| Use cross-fitting without loss correction (ablation) | Knowledge distillation · Teacher-overfitting settings | student error | Improved |
| Use cross-fitting without loss correction (ablation) | Knowledge distillation · Teacher-underfitting settings | student error | No clear effect |
Closest papers by meaning
- Bayesian Knowledge Distillation: A Bayesian Perspective of Distillation with Uncertainty Quantification (ICML 2024 · similarity 0.91)
- PLD: A Choice-Theoretic List-Wise Knowledge Distillation (NeurIPS 2025 · similarity 0.90)
- Can Students Beyond the Teacher? Distilling Knowledge from Teacher’s Bias (AAAI 2025 · similarity 0.90)
- Distilling What Matters: Confidence-Aware Selective Distillation for Large Language Models (NeurIPS 2026 · similarity 0.89)
- DistiLLM: Towards Streamlined Distillation for Large Language Models (ICML 2024 · similarity 0.89)
- ABKD: Pursuing a Proper Allocation of the Probability Mass in Knowledge Distillation via $\alpha$-$\beta$-Divergence (ICML 2025 · similarity 0.89)
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.
