Research map / Label-efficient and robust learning / Knowledge distillation
Explaining Knowledge Distillation by Quantifying the Knowledge
Xu Cheng, Zhefan Rao, Yilan Chen, Quanshi Zhang
Check 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 |
|---|---|---|---|
| Distill from the teacher instead of training from raw data | VGG-16 · CUB200-2011, FC1 | foreground visual concepts: 22.50 -> 29.57 | Improved |
| Distill from the teacher instead of training from raw data | VGG-16 · CUB200-2011, FC1 | foreground ratio λ: 0.68 -> 0.75 | Improved |
| Distill from the teacher instead of training from raw data | VGG-16 · CUB200-2011, FC2 | foreground visual concepts: 23.31 -> 32.92 | Improved |
| Distill from the teacher instead of training from raw data | VGG-16 · CUB200-2011, FC2 | foreground ratio λ: 0.67 -> 0.75 | Improved |
| Distill from the teacher instead of training from raw data | VGG-16 · ILSVRC-2013 DET, FC1 | foreground visual concepts: 36.47 -> 35.98 | Hurt |
| Distill from the teacher instead of training from raw data | VGG-16 · ILSVRC-2013 DET, FC1 | foreground ratio λ: 0.81 -> 0.84 | Improved |
| Distill from the teacher instead of training from raw data | VGG-16 · ILSVRC-2013 DET, FC2 | foreground visual concepts: 36.35 -> 42.74 | Improved |
| Distill from the teacher instead of training from raw data | VGG-16 · ILSVRC-2013 DET, FC2 | foreground ratio λ: 0.82 -> 0.82 | No clear effect |
| Use knowledge distillation rather than raw-data training | Compared student and baseline networks · CUB200-2011, ILSVRC-2013 DET, and Pascal VOC 2012 | learning speed and simultaneity (D_mean and D_std) | Mixed |
| Use knowledge distillation rather than raw-data training | Compared student and baseline networks · CUB200-2011, ILSVRC-2013 DET, and Pascal VOC 2012 | optimization-direction stability (ρ) | Mixed |
Closest papers by meaning
- Quantifying the Knowledge in a DNN to Explain Knowledge Distillation for Classification (TPAMI 2022 · similarity 0.92)
- A Gift From Knowledge Distillation: Fast Optimization, Network Minimization and Transfer Learning (CVPR 2017 · similarity 0.90)
- Harmonizing knowledge Transfer in Neural Network with Unified Distillation (ECCV 2024 · similarity 0.90)
- Knowledge Distillation: A Survey (IJCV 2021 · similarity 0.90)
- Neural Collapse Inspired Knowledge Distillation (AAAI 2025 · similarity 0.88)
- VkD: Improving Knowledge Distillation using Orthogonal Projections (CVPR 2024 · similarity 0.87)
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.
