Research map / Label-efficient and robust learning / Noisy labels
Wasserstein Distributional Normalization For Robust Distributional Certification of Noisy Labeled Data
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 |
|---|---|---|---|
| add Wasserstein distributional normalization (WDN) on uncertain samples | 9-layer CNN small-loss training · CIFAR-10/100 symmetric and asymmetric noise | test accuracy (%): WDN 87.40 vs JoCoR 85.73 at sym 20%; 82.89 vs 79.41 at sym 50% | Helped |
| combine WDN with co-teaching (WDNcot) (ablation) | WDN · CIFAR-10 asymmetric 45% | test accuracy (%): WDNcot 74.11 vs WDN 76.12 is lower, so WDNcot helps only vs co-teaching baseline (78.23 vs 74.11 is worse at sym 20%) | Helped |
| add WDN on uncertain samples | 9-layer CNN · CIFAR-10 open-set noise | test accuracy (%): 51.28 vs 47.73 (JoCoR) | Helped |
| add WDN on uncertain samples | ResNet50 pre-trained · Clothing1M | test accuracy (%): 73.49 vs DivideMix 74.76 | Mixed |
| add WDN (non-parametric SDE) instead of extra dual networks | training cost on CIFAR-10 · 5-epoch training time | training time overhead vs vanilla: +11.3% for WDN vs +201% for DivideMix | Helped |
| apply WDN from epoch 20 onward instead of never (uncertified baseline) (ablation) | 9-layer CNN training on CIFAR-10 sym 50% · CIFAR-10 symmetric 50% test accuracy trend | test accuracy | Helped |
| Wasserstein moving geodesic average of Gaussian parameters (alpha=0.2) (ablation) | WDN detour measure · CIFAR-10 noisy labels | qualitative stability of certified radius and accuracy | No clear effect |
Closest papers by meaning
- Wasserstein Gradient Boosting: A Framework for Distribution-Valued Supervised Learning (NeurIPS 2024 · similarity 0.87)
- Universal generalization guarantees for Wasserstein distributionally robust models (ICLR 2025 · similarity 0.87)
- Wasserstein Distributionally Robust Optimization through the Lens of Structural Causal Models and Individual Fairness (NeurIPS 2024 · similarity 0.86)
- Knowledge-Guided Wasserstein Distributionally Robust Optimization (ICML 2025 · similarity 0.86)
- Learning from Noisy Labels via Conditional Distributionally Robust Optimization (NeurIPS 2024 · similarity 0.86)
- Provable Robust Overfitting Mitigation in Wasserstein Distributionally Robust Optimization (ICLR 2025 · similarity 0.85)
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.