Research map / Label-efficient and robust learning / Noisy labels
CSOT: Curriculum and Structure-Aware Optimal Transport for Learning with Noisy Labels
Wanxing Chang, Ye Shi, Jingya Wang
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 curriculum constraints to classical OT (ablation) | classical OT · CIFAR-10/100 avg | accuracy: 78.07 -> 80.57 | Improved |
| add local coherent regularizers P and L (ablation) | CSOT w/o P and L · CIFAR-10/100 avg | accuracy: 80.57 -> 81.85 | Improved |
| add structure-aware regularization to OT (ablation) | classical OT · CIFAR-10/100 avg | accuracy: 78.07 -> 78.65 | Improved |
| curriculum on CIFAR-10 sym-0.9 (ablation) | classical OT · CIFAR-10 sym-0.9 | accuracy: 82.35 -> 89.50 | Improved |
| use CSOT instead of GMM for clean label identification (ablation) | GMM + L_sup · CIFAR-10/100 avg | accuracy: 62.04 -> 81.85 | Improved |
| use supervised loss L_sup instead of cross-entropy (ablation) | CSOT · CIFAR-10/100 avg | accuracy: 66.48 -> 81.85 | Improved |
| add semi-supervised loss on corrupted samples (ablation) | CSOT · CIFAR-10/100 avg | accuracy: 77.80 -> 81.85 | Improved |
| use CSOT relabeling instead of confidence thresholding 0.95 (ablation) | CSOT · CIFAR-10/100 avg | accuracy: 79.91 -> 81.85 | Improved |
| add self-supervised simsiam loss on corrupted set (ablation) | CSOT · CIFAR-100 sym-0.9 | accuracy: 45.56 -> 50.50 | Improved |
| efficient scaling iteration vs vanilla Dykstra solver (ablation) | CSOT solver · size (3000,3000) | time cost (s): 3.74 -> 0.99 | Improved |
| efficient scaling iteration vs vanilla Dykstra solver (ablation) | CSOT solver · size (1024,10) | time cost (s): 0.83 -> 0.82 | No clear effect |
| use CSOT instead of NCE | CSOT · CIFAR-100 sym-0.9 | accuracy: 41.1 -> 50.5 | 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.
- UniCon — UniCon: Combating Label Noise Through Uniform Selection and Contrastive Learning (CVPR 2022) · better on 8 of 9 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
- Revisiting Learning with Noisy Labels: Active Forgetting and Noise Suppression (CVPR 2026 · similarity 0.87)
- Robust Curriculum Learning: from clean label detection to noisy label self-correction (ICLR 2021 · similarity 0.87)
- Deconstructing the Failure of Ideal Noise Correction: A Three-Pillar Diagnosis (CVPR 2026 · similarity 0.87)
- Sample-wise Label Confidence Incorporation for Learning with Noisy Labels (ICCV 2023 · similarity 0.87)
- Label Noise: Ignorance Is Bliss (NeurIPS 2024 · similarity 0.86)
- Learning from Noisy Labels via Conditional Distributionally Robust Optimization (NeurIPS 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.
