Research map / Label-efficient and robust learning / Noisy labels
Learning From Noisy Labels With Distillation
Yuncheng Li, Jianchao Yang, Yale Song, Liangliang Cao, Jiebo Luo, Li-Jia Li
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 distillation from clean-data teacher | Baseline-Noisy · Sports | mAP: 50.7 -> 53.5 | Improved |
| add distillation from clean-data teacher | Baseline-Noisy · Species-I | mAP: 38.5 -> 41.6 | Improved |
| add distillation from clean-data teacher | Baseline-Noisy · Artifacts | mAP: 22.0 -> 26.0 | Improved |
| add knowledge graph guidance to distillation | Distillation · Sports | mAP: 53.5 -> 53.7 | Improved |
| add knowledge graph guidance to distillation | Distillation · Species-I | mAP: 41.6 -> 42.3 | Improved |
| add knowledge graph guidance to distillation | Distillation · Species-Y | mAP: 26.1 -> 25.2 | Hurt |
| add knowledge graph guidance to distillation | Distillation · Artifacts | mAP: 26.0 -> 26.0 | No clear effect |
| use distillation instead of label smoothing | noisy label learning · Artifacts | mAP: 22.9 -> 26.0 | Improved |
| use distillation instead of bootstrap | noisy label learning · Species-Y | mAP: 23.6 -> 26.1 | Improved |
| use distillation instead of importance re-weighting | noisy label learning · Artifacts | mAP: 24.8 -> 26.0 | Improved |
| use finetuning from clean model instead of training on noisy data only | Baseline-Noisy · Species-Y | mAP: 23.7 -> 22.2 | Hurt |
| use distillation instead of ensembling clean and noisy models | Baseline-Ensemble · Artifacts | mAP: 26.9 -> 26.0 | Hurt |
Closest papers by meaning
- DualGraph: A Graph-Based Method for Reasoning About Label Noise (CVPR 2021 · similarity 0.88)
- Learning Discriminative Dynamics with Label Corruption for Noisy Label Detection (CVPR 2024 · similarity 0.88)
- A Label is Worth A Thousand Images in Dataset Distillation (NeurIPS 2024 · similarity 0.88)
- Learning with Structural Labels for Learning with Noisy Labels (CVPR 2024 · similarity 0.87)
- Rethinking Self-Distillation: Label Averaging and Enhanced Soft Label Refinement with Partial Labels (ICLR 2025 · similarity 0.87)
- Adaptive Integration of Partial Label Learning and Negative Learning for Enhanced Noisy Label Learning (AAAI 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.
