Research map / Label-efficient and robust learning / Noisy labels
Sample Selection with Uncertainty of Losses for Learning with Noisy Labels
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 |
|---|---|---|---|
| Use CNLCU-S with uncertainty-aware sample selection | noisy-label learning · Clothing1M Best | test accuracy (%): +1.28% over the best baseline JoCor | Helped |
| Use CNLCU-S with uncertainty-aware sample selection | noisy-label learning · Clothing1M Last | test accuracy (%): +1.01% over JoCor | Helped |
| Use uncertainty-aware CNLCU selection | noisy-label learning · synthetic balanced datasets with multiple noise types | test accuracy | Helped |
| Add conservative selection of less-selected, large-loss examples (ablation) | CNLCU-S · MNIST and F-MNIST noisy-label settings | test accuracy (%): 98.02 -> 98.82 (MNIST Sym. 20%); 72.93 -> 83.14 (F-MNIST Asym. 40%) | Helped |
| Add robust multi-iteration estimation for small-loss examples (ablation) | CNLCU-S · MNIST and F-MNIST noisy-label settings | test accuracy (%): 98.15 -> 98.82 (MNIST Sym. 20%); 80.99 -> 83.14 (F-MNIST Asym. 40%) | Helped |
| Replace non-robust mean with robust mean estimation (ablation) | Co-teaching-M · MNIST noisy-label settings | test accuracy (%): 97.72 -> 98.82 (Sym. 20%) | Helped |
| Add conservative selection of less-selected, large-loss examples (ablation) | CNLCU-H · MNIST and F-MNIST noisy-label settings | test accuracy (%): 98.06 -> 98.70 (MNIST Sym. 20%); 71.35 -> 82.69 (F-MNIST Asym. 40%) | Helped |
| Add robust multi-iteration estimation for small-loss examples (ablation) | CNLCU-H · MNIST and F-MNIST noisy-label settings | test accuracy (%): 98.19 -> 98.70 (MNIST Sym. 20%); 80.73 -> 82.69 (F-MNIST Asym. 40%) | Helped |
| Vary the fixed-length loss time interval within the tested ranges (ablation) | CNLCU-S and CNLCU-H · MNIST and F-MNIST noisy-label settings | test accuracy | No clear effect |
| Use CNLCU to encourage selection of underrepresented examples | noisy-label learning · IM-MNIST and IM-F-MNIST | test accuracy: nearly 30% lead in IM-MNIST Asym. 10%; about 5% lead on IM-F-MNIST | Helped |
| Use conservative uncertainty-based selection | CNLCU · IM-MNIST and IM-F-MNIST | selected ratio (%): 0.60 vs 0.10 (IM-MNIST 10% noise, CNLCU-S vs MentorNet) | Helped |
Closest papers by meaning
- Enhanced Sample Selection with Confidence Tracking: Identifying Correctly Labeled Yet Hard-to-Learn Samples in Noisy Data (AAAI 2025 · similarity 0.89)
- Enhancing Noise-Robust Losses for Large-Scale Noisy Data Learning (AAAI 2025 · similarity 0.88)
- Regroup Median Loss for Combating Label Noise (AAAI 2024 · similarity 0.88)
- On the Role of Label Noise in the Feature Learning Process (ICML 2025 · similarity 0.87)
- Enhancing Sample Selection Against Label Noise by Cutting Mislabeled Easy Examples (NeurIPS 2025 · similarity 0.87)
- Learning from Noisy Labels via Conditional Distributionally Robust Optimization (NeurIPS 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.