Research map / Label-efficient and robust learning / Knowledge distillation
Learning From Multiple Experts: Self-paced Knowledge Distillation for Long-tailed Classification
Liuyu Xiang, Guiguang Ding, Jungong Han
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 ordinary knowledge distillation from experts to instance-level sampling (ablation) | plain ResNet-10 student · ImageNet-LT | All-class accuracy: 20.9 -> 32.2 | Improved |
| add ordinary knowledge distillation from experts to class-level sampling (ablation) | class-sampling student · ImageNet-LT | All-class accuracy: 32.6 -> 35.8 | Improved |
| switch from instance-level to class-level random sampling (ablation) | plain ResNet-10 · ImageNet-LT | All-class accuracy: 20.9 -> 32.6 | Improved |
| add Self-paced Expert Selection to distillation (ablation) | Cls.Samp.+KD · ImageNet-LT | All-class accuracy: 35.8 -> 37.1 | Improved |
| add Curriculum Instance Selection (ablation) | Cls.Samp.+KD+SpES · ImageNet-LT | Few-shot accuracy: 16.5 -> 17.5 | Improved |
| use convex instead of linear instance schedule (ablation) | LFME · ImageNet-LT | All-class accuracy: 37.2 -> 36.9 | Hurt |
| use concave instead of linear instance schedule (ablation) | LFME · ImageNet-LT | All-class accuracy: 37.2 -> 37.1 | No clear effect |
| set alpha to 1.0, making self-paced selection ordinary distillation (ablation) | LFME · ImageNet-LT | accuracy | Hurt |
| add LFME to OLTR second stage | OLTR · ImageNet-LT | All-class accuracy: 35.6 -> 38.8 | Improved |
| add LFME to Focal Loss | Focal Loss · ImageNet-LT | All-class accuracy: 30.5 -> 37.3 | Improved |
| add LFME to OLTR | OLTR · Places-LT | All-class accuracy: 35.9 -> 36.2 | Improved |
| add LFME to Focal Loss | Focal Loss · Places-LT | All-class accuracy: 34.6 -> 35.2 | Improved |
Closest papers by meaning
- Knowledge Distillation by On-the-Fly Native Ensemble (NeurIPS 2018 · similarity 0.88)
- Densely Guided Knowledge Distillation Using Multiple Teacher Assistants (ICCV 2021 · similarity 0.87)
- Breaking Long-Tailed Learning Bottlenecks: A Controllable Paradigm with Hypernetwork-Generated Diverse Experts (NeurIPS 2024 · similarity 0.87)
- Efficient Knowledge Distillation from Model Checkpoints (NeurIPS 2022 · similarity 0.86)
- Distilling Balanced Knowledge from a Biased Teacher (CVPR 2026 · similarity 0.86)
- DeepKD: A Deeply Decoupled and Denoised Knowledge Distillation Trainer (NeurIPS 2025 · similarity 0.85)
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.
