Research map / Label-efficient and robust learning / Noisy labels
Open-set Label Noise Can Improve Robustness Against Inherent Label Noise
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 ODNL to standard training | WRN-40-2 on CIFAR-10 · CIFAR-10 Sym-50% | test accuracy (%): 69.48±0.58 -> 82.50±0.58 | Helped |
| Add ODNL to standard training | WRN-40-2 on CIFAR-10 · CIFAR-10 Asymmetric | test accuracy (%): 78.34±0.96 -> 90.00±0.35 | Helped |
| Add ODNL to JoCoR | WRN-40-2 on CIFAR-10 · CIFAR-10 Sym-50% | test accuracy (%): 88.36±0.64 -> 91.17±0.52 | Helped |
| Add ODNL to standard training | WRN-40-2 on CIFAR-100 · CIFAR-100 Dependent | test accuracy (%): 53.94±0.52 -> 62.45±0.41 | Helped |
| Add ODNL to DivideMix | ResNet-50 pretrained on Clothing1M · Clothing1M best | test accuracy (%): 74.32 -> 74.89 | Helped |
| Use ODNL rather than MSP alone | CIFAR-10 model with symmetric-40% training noise · OOD test sets | FPR95: 92.01 -> 13.43 | Helped |
| Use ODNL rather than OE | CIFAR-10 model with symmetric-40% training noise · OOD test sets | AUROC: 94.24 -> 96.67 | Helped |
| Replace open-set auxiliary data with closed-set auxiliary data (ablation) | Standard training on CIFAR-10 with inherent label noise · CIFAR-10 noisy-label robustness | test accuracy | Hurt |
| Use fixed random auxiliary labels instead of changing them dynamically (ablation) | ODNL · Noisy-label robustness | test accuracy | Hurt |
| Increase the size of fixed-label open-set auxiliary data (ablation) | Standard training on CIFAR-10 · CIFAR-10 with inherent noisy labels | test accuracy | Helped |
| Use more realistic open-set images instead of Gaussian noise (ablation) | ODNL on CIFAR-10 symmetric-40% noise · CIFAR-10 symmetric-40% | test accuracy: 80.24% -> 82.03% | Helped |
| Vary the number of classes in the auxiliary CIFAR-100 subset (ablation) | ODNL on CIFAR-10 symmetric-40% noise · CIFAR-10 symmetric-40% | test accuracy: 81.56% | No clear effect |
Closest papers by meaning
- Unlocking the Power of Open Set: A New Perspective for Open-Set Noisy Label Learning (AAAI 2024 · similarity 0.90)
- Learning with Open-world Noisy Data via Class-independent Margin in Dual Representation Space (AAAI 2025 · similarity 0.89)
- Label Noise: Ignorance Is Bliss (NeurIPS 2024 · similarity 0.89)
- Handling Label Noise via Instance-Level Difficulty Modeling and Dynamic Optimization (NeurIPS 2025 · similarity 0.89)
- Mitigating Label Noise through Data Ambiguation (AAAI 2024 · similarity 0.88)
- On the Role of Label Noise in the Feature Learning Process (ICML 2025 · 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.