Research map / Label-efficient and robust learning / Noisy labels
Learning From Noisy Labels With Decoupled Meta Label Purifier
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| apply DMLP-purified labels before DivideMix training | DivideMix · CIFAR-10/100 symmetric noise, 20%–90% | best test accuracy: 94.6 -> 94.5 (CIFAR-10 80%); 60.2 -> 68.6 (CIFAR-100 80%) | Helped |
| apply DMLP-purified labels before DivideMix training | DivideMix · Clothing1M testset | Top-1 accuracy: 74.76 -> 78.23 | Helped |
| add IPC to EAC (ablation) | DMLP-Naive · CIFAR-10 90% symmetric noise | best test accuracy: 67.4 -> 92.8 | Helped |
| add EAC to IPC (ablation) | DMLP-Naive · CIFAR-10 80% symmetric noise | best test accuracy: 79.9 -> 93.5 | Helped |
| add IPC to EAC (ablation) | DMLP-Naive · CIFAR-10 20% symmetric noise | best test accuracy: 93.7 -> 94.7 | Helped |
| add EAC to IPC (ablation) | DMLP-Naive · Clothing1M | accuracy: 76.8 -> 77.7 | Helped |
| replace ResNet-18 MoCo features with ResNet-50 SimCLR features (ablation) | DMLP-Naive · CIFAR-10 symmetric noise | best test accuracy: 93.8 -> 94.7 (20%); 90.4 -> 92.8 (90%) | Helped |
| reduce validation set size to 10% (ablation) | DMLP-Naive · Clothing1M | accuracy: 77.31 -> 75.50 | Hurt |
| use DMLP-DivideMix rather than semi-supervised baselines | CIFAR-10/100 at 100% symmetric noise · CIFAR-10 and CIFAR-100 | accuracy: 88.2 -> 91.7 (CIFAR-10); 57.7 -> 60.1 (CIFAR-100) | Helped |
Closest papers by meaning
- Learning to Purify Noisy Labels via Meta Soft Label Corrector (AAAI 2021 · similarity 0.93)
- Revisiting Interpolation for Noisy Label Correction (AAAI 2025 · similarity 0.87)
- Adaptive Integration of Partial Label Learning and Negative Learning for Enhanced Noisy Label Learning (AAAI 2024 · similarity 0.87)
- Learning with Structural Labels for Learning with Noisy Labels (CVPR 2024 · similarity 0.86)
- Mitigating Label Noise through Data Ambiguation (AAAI 2024 · similarity 0.86)
- Learning Discriminative Dynamics with Label Corruption for Noisy Label Detection (CVPR 2024 · similarity 0.86)
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.