Research map / Label-efficient and robust learning / Noisy labels
Combating Label Noise with a General Surrogate Model for Sample Selection
Chao Liang, Linchao Zhu, Humphrey Shi, Yi Yang
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 CLIP-based clean sample selection and new loss | DivideMix · WebVision | top-1 accuracy: 77.32 -> 79.08 | Improved |
| add CLIP-based selection and loss | DivideMix · CIFAR-10 sym 90% | accuracy: 75.4 -> 89.2 | Improved |
| add CLIP-based selection and loss | DivideMix · CIFAR-100 sym 90% | accuracy: 31.0 -> 45.7 | Improved |
| add CLIP-based selection and loss | DivideMix · CIFAR-10 asym 40% | accuracy: 92.1 -> 95.1 | Improved |
| add CLIP-based selection and loss | DivideMix · Clothing-1M | accuracy: 74.76 -> 74.84 | No clear effect |
| add CLIP-based selection | InstanceGM · Red Mini-ImageNet 80% | accuracy: 39.62 -> 45.65 | Improved |
| select samples by CLIP prediction confidence instead of GMM small loss (ablation) | sampling strategy · WebVision | top-1 accuracy: 77.36 -> 78.32 | Improved |
| select by CLIP prompt consistency instead of prediction confidence (ablation) | sampling strategy · WebVision | top-1 accuracy: 78.32 -> 77.36 | Hurt |
| add focal loss instead of cross-entropy (ablation) | classifier training · WebVision | top-1 accuracy: 77.76 -> 78.32 | Improved |
| add focal loss to full loss (ablation) | noise-aware balanced margin loss · WebVision | top-1 accuracy: 78.20 -> 79.08 | Improved |
| add balanced margin (ablation) | noise-aware balanced margin loss · WebVision | top-1 accuracy: 78.68 -> 79.08 | Improved |
| add noise-aware margin (ablation) | noise-aware balanced margin loss · WebVision | top-1 accuracy: 78.76 -> 79.08 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DivideMix — DivideMix: Learning with Noisy Labels as Semi-supervised Learning (ICLR 2020) · better on 10 of 10 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
- On the Role of Label Noise in the Feature Learning Process (ICML 2025 · similarity 0.88)
- Optimized Gradient Clipping for Noisy Label Learning (AAAI 2025 · similarity 0.88)
- A CLIP-Powered Framework for Robust and Generalizable Data Selection (ICLR 2025 · similarity 0.87)
- Enhancing Sample Selection Against Label Noise by Cutting Mislabeled Easy Examples (NeurIPS 2025 · similarity 0.87)
- Enhanced Sample Selection with Confidence Tracking: Identifying Correctly Labeled Yet Hard-to-Learn Samples in Noisy Data (AAAI 2025 · similarity 0.87)
- Mitigating Label Noise through Data Ambiguation (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.
