Research map / Label-efficient and robust learning / Knowledge distillation
Improving Knowledge Distillation Under Unknown Covariate Shift Through Confidence-Guided Data Augmentation
Niclas Popp, Kevin Alexander Laube, Matthias Hein, Lukas Schott
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 ConfiG teacher-student-disagreement augmentations | KD ViT-T student · CelebA-HQ | worst group accuracy (%): 53.44 -> 88.15 | Improved |
| ConfiG vs best baseline ActGen | KD ViT-T student · CelebA-HQ | worst group accuracy (%): 75.68 -> 88.15 | Improved |
| ConfiG vs Diff-Mix | KD ViT-T student · SpuCo Birds | worst group accuracy (%): 32.13 -> 39.50 | Improved |
| ConfiG vs Diff-Mix | KD ViT-T student · BAR | sample mean accuracy (%): 59.28 -> 58.66 | Hurt |
| remove inter-class mixing, keep latent optimization (ablation) | ConfiG · SpuCo Birds | worst group accuracy (%): 21.63 -> 39.50 | Improved |
| add latent optimization to inter-class mixing (ablation) | ConfiG · BAR | sample mean accuracy (%): 44.09 -> 58.66 | Improved |
| add latent optimization to inter-class mixing (ablation) | ConfiG · CelebA-HQ | worst group accuracy (%): 86.99 -> 88.15 | Improved |
| latent optimization alone (ablation) | ConfiG · CelebA-HQ | sample mean accuracy (%): 93.98 -> 93.35 | Hurt |
| add stronger standard augmentation (AutoAugment) to ConfiG (ablation) | ConfiG · SpuCo Birds | worst group accuracy (%): 39.50 -> 50.53 | Improved |
| double the number of ConfiG augmentations per image (ablation) | ConfiG · SpuCo Birds | worst group accuracy (%): 39.50 -> 46.90 | Improved |
| double Diff-Mix augmentations (28 to 56) (ablation) | Diff-Mix · SpuCo Birds | worst group accuracy (%): 32.13 -> 34.53 | Improved |
| use weaker teacher (ablation) | ConfiG · CelebA-HQ / BAR | accuracy | Hurt |
Closest papers by meaning
- Small Scale Data-Free Knowledge Distillation (CVPR 2024 · similarity 0.87)
- D^4: Dataset Distillation via Disentangled Diffusion Model (CVPR 2024 · similarity 0.87)
- DKDM: Data-Free Knowledge Distillation for Diffusion Models with Any Architecture (CVPR 2025 · similarity 0.87)
- De-confounded Data-free Knowledge Distillation for Handling Distribution Shifts (CVPR 2024 · similarity 0.87)
- Your Student is Better Than Expected: Adaptive Teacher-Student Collaboration for Text-Conditional Diffusion Models (CVPR 2024 · similarity 0.86)
- Unlocking Dataset Distillation with Diffusion Models (NeurIPS 2025 · 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.
