Research map / Label-efficient and robust learning / Noisy labels
Toward Robustness against Label Noise in Training Deep Discriminative Neural Networks
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 auxiliary-distribution regularization while inferring clean labels | CNN-CRF on CIFAR-10 · CIFAR-10, 50% noise | Prediction accuracy (%): 80.0 -> 84.3 | Helped |
| Use the model's inferred labels to recover noisy training labels | CNN-CRF on CIFAR-10 · CIFAR-10, 50% noise | Recovery accuracy (%): 83.1 -> 88.1 | Helped |
| Add hidden variables and their pairwise terms (ablation) | COCO Caption Labels, ResNet-50 Small · COCO Caption Labels, ResNet-50 Small | mAP (%): 64.26 -> 65.73 | Helped |
| Remove the x-y link from CRF with hidden (ablation) | COCO Caption Labels, ResNet-50 · COCO Caption Labels, ResNet-50 Small | mAP (%): 65.73 -> 66.61 | Helped |
| Set alpha to 0 while removing the x-y link (ablation) | COCO Caption Labels, ResNet-50 · COCO Caption Labels, ResNet-50 Small | mAP (%): 66.61 -> 48.53 | Hurt |
| Remove pairwise terms (ablation) | CNN-CRF on COCO Caption Labels · COCO Caption Labels, ResNet-50 Small | mAP (%): 63.67 vs 66.61 | Hurt |
| Add hidden-variable pairwise terms (ablation) | CNN-CRF on COCO Caption Labels · COCO Caption Labels, ResNet-50 Large | mAP (%): 73.23 -> 74.04 | Helped |
| Remove the x-y link from CRF with hidden (ablation) | CNN-CRF on COCO Caption Labels · COCO Caption Labels, ResNet-50 Large | mAP (%): 74.04 -> 75.00 | Helped |
| Remove the x-y link and set alpha to 0 (ablation) | CNN-CRF on COCO Caption Labels · COCO Caption Labels, ResNet-50 Large | mAP (%): 75.00 -> 56.53 | Hurt |
| Remove hidden variables and their pairwise terms (ablation) | CNN-CRF on COCO Caption Labels · COCO Caption Labels, ResNet-50 Small | mAP (%): 65.73 -> 64.26 | Hurt |
| Combine image-conditioned inference with the auxiliary label distribution | CNN-CRF, ResNet-50 Large · COCO noisy training set | mAP (%): 49.4% -> 69.4% | Helped |
Closest papers by meaning
- Distilling Effective Supervision From Severe Label Noise (CVPR 2020 · similarity 0.92)
- Coresets for Robust Training of Deep Neural Networks against Noisy Labels (NeurIPS 2020 · similarity 0.91)
- Joint Optimization Framework for Learning With Noisy Labels (CVPR 2018 · similarity 0.91)
- FINE Samples for Learning with Noisy Labels (NeurIPS 2021 · similarity 0.90)
- On the Role of Label Noise in the Feature Learning Process (ICML 2025 · similarity 0.90)
- Learning with Structural Labels for Learning with Noisy Labels (CVPR 2024 · similarity 0.89)
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.