Research map / Label-efficient and robust learning / Noisy labels
Distilling Effective Supervision From Severe Label Noise
Check 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 meta-estimated exemplar weights and pseudo labels with the trusted probe set | Ours · CIFAR100 uniform noise 0.8 | accuracy: 48.2 -> 75.5 ± 0.2 | Helped |
| Add meta re-labeling (λ) (ablation) | Full method · CIFAR100 noise 0.8 | accuracy: 74.04 -> 75.42 | Helped |
| Add meta re-labeling (λ) (ablation) | Full method · CIFAR100 noise 0.4 | accuracy: 81.05 -> 79.96 | Hurt |
| Add KL consistency loss (ablation) | Full method · CIFAR100 noise 0.8 | accuracy: 61.81 -> 75.42 | Helped |
| Add KL consistency loss (ablation) | Full method · CIFAR100 noise 0.4 | accuracy: 78.06 -> 79.96 | Helped |
| Add mixup loss for probe-data anchoring (ablation) | Full method · CIFAR100 noise 0.8 | accuracy: 54.76 -> 75.42 | Helped |
| Add mixup loss for probe-data anchoring (ablation) | Full method · CIFAR100 noise 0.4 | accuracy: 73.63 -> 79.96 | Helped |
| Use learned policy augmentation (ablation) | Full method · CIFAR100 noise 0.8 | accuracy: 72.69 -> 75.42 | Helped |
| Use learned policy augmentation (ablation) | Full method · CIFAR100 noise 0.4 | accuracy: 79.16 -> 79.96 | Helped |
| Remove mixup when using 10 trusted samples per class (ablation) | Full method · CIFAR100 noise 0.8 | accuracy: 72.7 -> 78.4 | Hurt |
| Remove mixup when using 1 trusted sample per class (ablation) | Full method · CIFAR100 noise 0.8 | accuracy: 47.1 vs 62.1 | Hurt |
| Add KL consistency to MixMatch | MixMatch · CIFAR100, 10 labeled samples per class | accuracy: 34.5 -> 57.6 | Helped |
Closest papers by meaning
- Using Trusted Data to Train Deep Networks on Labels Corrupted by Severe Noise (NeurIPS 2018 · similarity 0.93)
- Toward Robustness against Label Noise in Training Deep Discriminative Neural Networks (NeurIPS 2017 · similarity 0.92)
- Learning with Structural Labels for Learning with Noisy Labels (CVPR 2024 · similarity 0.89)
- Handling Label Noise via Instance-Level Difficulty Modeling and Dynamic Optimization (NeurIPS 2025 · similarity 0.89)
- On the Role of Label Noise in the Feature Learning Process (ICML 2025 · similarity 0.89)
- Mitigating Label Noise through Data Ambiguation (AAAI 2024 · similarity 0.88)
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.