Research map / Label-efficient and robust learning / Noisy labels
CMW-Net: Learning a Class-Aware Sample Weighting Mapping for Robust Deep Learning
Jun Shu, Xiang Yuan, Deyu Meng, Zongben Xu
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 class-aware weighting (CMW-Net) in place of single class-agnostic MW-Net | MW-Net · CIFAR-10-LT (imbalance 200) | top-1 error (%): 32.80 -> 27.80 | Improved |
| add class-aware weighting (CMW-Net) in place of MW-Net | MW-Net · CIFAR-100-LT (imbalance 100) | top-1 error (%): 58.39 -> 55.25 | Improved |
| add CMW-Net on top of LDAM loss | LDAM · CIFAR-10-LT (imbalance 100) | top-1 error (%): 26.65 -> 19.95 | Improved |
| add CMW-Net on top of SADE | SADE · CIFAR-10-LT (imbalance 100) | top-1 error (%): 16.78 -> 16.04 | Improved |
| use class-aware weighting in a balanced dataset (imbalance factor 1) | MW-Net · CIFAR-10-LT (imbalance 1) | top-1 error (%): 7.19 -> 8.30 | Hurt |
| use CMW-Net instead of MW-Net under asymmetric noise | MW-Net · CIFAR-10 asymmetric noise 0.8 | test accuracy (%): 58.29 -> 87.50 | Improved |
| add soft label supervision (pseudo-labels) to CMW-Net (ablation) | CMW-Net · CIFAR-10 symmetric noise 0.8 | test accuracy (%): 64.80 -> 92.10 | Improved |
| add soft label supervision to CMW-Net (ablation) | CMW-Net · CIFAR-100 asymmetric noise 0.8 | test accuracy (%): 17.41 -> 18.34 | Improved |
| add soft label supervision to CMW-Net (ablation) | CMW-Net · Animal-10N | test accuracy (%): 80.9 -> 84.7 | Improved |
| add soft label supervision to CMW-Net (ablation) | CMW-Net · CIFAR-10 feature-dependent Type-I 35% | test accuracy (%): 82.27 -> 84.23 | Improved |
| add self-supervised warm-up and consistency regularization (SL+) (ablation) | CMW-Net-SL · CIFAR-100 symmetric noise 0.9 | test accuracy (%): 21.2 -> 64.6 | Improved |
| use CMW-Net-SL instead of DivideMix at extreme symmetric noise | DivideMix · CIFAR-10 symmetric noise 0.8 | test accuracy (%): 92.90 -> 92.10 | Hurt |
Later papers that compared with this method
1 later accepted paper reports CMW-Net as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Dynamic Loss for Robust Learning (TPAMI 2023 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- DivideMix — DivideMix: Learning with Noisy Labels as Semi-supervised Learning (ICLR 2020) · better on 2 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Meta-Weight-Net: Learning an Explicit Mapping For Sample Weighting (NeurIPS 2019 · similarity 0.94)
- Delving into Sample Loss Curve to Embrace Noisy and Imbalanced Data (AAAI 2022 · similarity 0.88)
- Foster Adaptivity and Balance in Learning with Noisy Labels (ECCV 2024 · similarity 0.87)
- Gradient-Weight Alignment as a Train-Time Proxy for Generalization in Classification Tasks (NeurIPS 2025 · similarity 0.86)
- Revive Re-weighting in Imbalanced Learning by Density Ratio Estimation (NeurIPS 2024 · similarity 0.85)
- Improving Generalization via Meta-Learning on Hard Samples (CVPR 2024 · similarity 0.84)
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.
