Research map / Label-efficient and robust learning / Noisy labels
Robustness to Label Noise Depends on the Shape of the Noise Distribution
Diane Oyen, Michal Kucer, Nick Hengartner, Har Simrat Singh
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 |
|---|---|---|---|
| feature-dependent (GapMax hash-based) instead of class-dependent noise (ablation) | ResNet-32 baseline · CIFAR10 eps=0.2, s=1 | clean accuracy: 78.15 -> 76.92 | Hurt |
| feature-dependent instead of class-dependent noise (ablation) | ResNet-32 baseline · CIFAR10 eps=0.6, s=5 | clean accuracy: 48.70 -> 40.70 | Hurt |
| feature-dependent instead of class-dependent noise (ablation) | ResNet-32 baseline · CIFAR100 eps=0.4, s=1 | clean accuracy: 42.24 -> 37.17 | Hurt |
| add CleanLab | ResNet-32 · CIFAR10 feature noise eps=0.4, s=1 | clean accuracy: 59.86 -> 82.47 | Improved |
| add MixUp | ResNet-32 · CIFAR100 feature noise eps=0.4, s=1 | clean accuracy: 37.17 -> 47.73 | Improved |
| add CleanLab | ResNet-32 · CIFAR100 class noise eps=0.4, s=1 | clean accuracy: 42.24 -> 32.84 | Hurt |
| add Co-Teaching | ResNet-32 · CIFAR10 class noise eps=0.6, s=1 | clean accuracy: 35.26 -> 22.80 | Hurt |
| add SCE | ResNet-32 · CIFAR10 class noise eps=0.4, s=1 | clean accuracy: 58.94 -> 57.42 | No clear effect |
| add MixUp | ResNet-32 · CIFAR10 class noise eps=0.6, s=1 | clean accuracy: 35.26 -> 12.27 | Hurt |
| add CleanLab or CoTeaching at higher noise levels | 2-hidden-layer NN · synthetic 2D 10-class feature noise | clean accuracy | Mixed |
| uniform-x feature noise (constant flip probability) vs class-dependent noise (ablation) | 2-hidden-layer NN · synthetic 2D | clean accuracy | No clear effect |
| GapMax worst-case noise vs other noise types (ablation) | 2-hidden-layer NN · synthetic 2D | clean accuracy | Hurt |
Closest papers by meaning
- Learning with Feature-Dependent Label Noise: A Progressive Approach (ICLR 2021 · similarity 0.89)
- Label Noise: Ignorance Is Bliss (NeurIPS 2024 · similarity 0.87)
- On the Role of Label Noise in the Feature Learning Process (ICML 2025 · similarity 0.87)
- Leveraging Data Symmetries to Select an Optimal Subset of Training Data under Label Noise (NeurIPS 2026 · similarity 0.87)
- Estimating Noisy Class Posterior with Part-level Labels for Noisy Label Learning (CVPR 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.
