Research map / Optimization / Implicit bias and generalization
Robust and On-the-fly Dataset Denoising for Image Classification
Jiaming Song, Yann Dauphin, Michael Auli, Tengyu Ma
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 |
|---|---|---|---|
| Add ODD (remove high-loss examples at epoch E) | WRN-28-10 ERM · CIFAR-10 40% noise | accuracy: 84.4 ± 0.5 -> 92.8 ± 0.2 | Improved |
| Add ODD | WRN-28-10 ERM · CIFAR-100 40% noise | accuracy: 55.7 ± 0.5 -> 72.4 ± 0.4 | Improved |
| Add mixup on top of ODD | ODD · CIFAR-10 40% noise | accuracy: 92.8 ± 0.2 -> 95.5 ± 0.2 | Improved |
| Add mixup on top of ODD | ODD · CIFAR-100 20% noise | accuracy: 77.2 ± 0.1 -> 79.1 ± 0.1 | Improved |
| Add ODD on clean data | WRN-28-10 ERM · CIFAR-10 0% noise | accuracy: 96.3 ± 0.1 -> 96.2 ± 0.1 | No clear effect |
| Add ODD | ResNet-152 ERM · ImageNet 40% noise | top-1 accuracy: 61.2 -> 74.8 | Improved |
| Add ODD on clean data | ResNet-152 ERM · ImageNet 0% noise | top-1 accuracy: 78.7 -> 78.7 | No clear effect |
| Add ODD to non-homogeneous class ratios | ERM · CIFAR-50 (c=70) | accuracy: 77.5 ± 0.1 -> 78.1 ± 0.1 | Improved |
| Add ODD on WebVision | ERM · WebVision val | top-1 accuracy: 69.7 -> 74.6 | Improved |
| Add ODD on noisy Clothing1M | ERM · Clothing1M noisy | accuracy: 68.9 -> 73.5 | Improved |
| Raise percentile p from 1 to 30 (ablation) | ODD · WebVision val | top-1 accuracy: 74.01 -> 74.62 | Improved |
| Vary percentile p (ablation) | ODD · ImageNet noisy | accuracy | No clear effect |
Closest papers by meaning
- Late Stopping: Avoiding Confidently Learning from Mislabeled Examples (ICCV 2023 · similarity 0.88)
- Coresets for Robust Training of Deep Neural Networks against Noisy Labels (NeurIPS 2020 · similarity 0.87)
- Enhancing Sample Selection Against Label Noise by Cutting Mislabeled Easy Examples (NeurIPS 2025 · similarity 0.86)
- MaxSup: Overcoming Representation Collapse in Label Smoothing (NeurIPS 2025 · similarity 0.86)
- Enhancing Noise-Robust Losses for Large-Scale Noisy Data Learning (AAAI 2025 · similarity 0.85)
- Revisiting Interpolation for Noisy Label Correction (AAAI 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.
