Research map / Label-efficient and robust learning / Knowledge distillation
Shadow Knowledge Distillation: Bridging Offline and Online Knowledge Transfer
Lujun Li, ZHE JIN
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 reversed distillation from student to teacher (KD to KD-dagger) | KD · CIFAR-100 R110->R20 | Top-1: 70.66 -> 71.76 | Improved |
| Add reversed distillation to online DML (DML-dagger to DML) | DML · CIFAR-100 R110->R20 | Top-1: 70.55 -> 71.52 | Improved |
| Online versus offline teacher training fashion (DML-dagger vs KD) | KD · CIFAR-100 R110->R20 | Top-1: 70.55 vs 70.66 | No clear effect |
| Reversed distillation reduces teacher-student gap | KD · CIFAR-100 R110->R20 | T-S gap (KL): 1.12 -> 0.66 | Improved |
| Replace KD-dagger with SHAKE shadow heads on shared proxy teacher | KD-dagger · CIFAR-100 R110->R20 | Time: x4.29 -> x1.28 | Improved |
| Update only head versus adding adapters versus proxy teacher | KD-dagger · CIFAR-100 R110->R20 | Top-1: 71.05 -> 71.82 | Improved |
| Share backbone in proxy teacher (alone to share) (ablation) | SHAKE · CIFAR-100 R110->R20 | Top-1: 71.82 -> 72.02 | Improved |
| Multi-teacher (multi-shadow) setting (ablation) | SHAKE · CIFAR-100 R110->R20 | Top-1: 72.02 -> 72.35 | Improved |
| Add SHAKE to two teachers (SHAKE-dagger) | SHAKE · ImageNet R34->R18 | Top-1: 72.07 -> 72.53 | Improved |
| Replace KD with SHAKE | KD · ImageNet R34->R18 | Top-1: 70.66 -> 72.07 | Improved |
| Replace KD with SHAKE | KD · CIFAR-100 ResNet32x4->ResNet8x4 | Top-1: 73.33 -> 77.35 | Improved |
| Replace KD with SHAKE | KD · ImageNet RegNetY-16GF->ViT-T | Top-1: 72.20 -> 75.22 | Improved |
Closest papers by meaning
- Student Customized Knowledge Distillation: Bridging the Gap Between Student and Teacher (ICCV 2021 · similarity 0.90)
- Cooperative Knowledge Distillation: A Learner Agnostic Approach (AAAI 2024 · similarity 0.90)
- Learning Student-Friendly Teacher Networks for Knowledge Distillation (NeurIPS 2021 · similarity 0.90)
- All You Need in Knowledge Distillation Is a Tailored Coordinate System (AAAI 2025 · similarity 0.89)
- Less or More From Teacher: Exploiting Trilateral Geometry For Knowledge Distillation (ICLR 2024 · similarity 0.89)
- Progressively Knowledge Distillation via Re-parameterizing Diffusion Reverse Process (AAAI 2024 · 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.
