Research map / Label-efficient and robust learning / Noisy labels
Scalable Penalized Regression for Noise Detection in Learning With Noisy Labels
Yikai Wang, Xinwei Sun, Yanwei Fu
Check 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 |
|---|---|---|---|
| Use SPR noisy-sample selection and train on selected clean data | C2F2 · MNIST, symmetric noise rates 0.2–0.8 | accuracy: 99.3 vs 99.2; 99.2 vs 99.2; 99.2 vs 98.9; 98.7 vs 98.0 | Improved |
| Use SPR noisy-sample selection with the full training framework | ResNet-18 · CIFAR-10, symmetric noise rates 0.2–0.8 | accuracy: 93.2 vs 90.2; 91.0 vs 87.2; 82.7 vs 80.5; 64.1 vs 45.7 | Improved |
| Use SPR noisy-sample selection with the full training framework | ResNet-18 · CIFAR-10, asymmetric noise rates 0.2–0.4 | accuracy: 92.8 vs 90.5; 91.3 vs 89.7; 89.0 vs 87.9 | Improved |
| Use SPR noisy-sample selection | VGG19-BN · ANIMAL10 | accuracy: 86.8 vs 84.1 | Improved |
| Use SPR noisy-sample selection | Inception-ResNet · WebVision | accuracy: 78.12 vs 77.32 | Improved |
| Add SPR detection and train with CE only on detected clean data (ablation) | CE training · CIFAR-10, symmetric noise 0.4 | accuracy: 65.5 -> 80.4 | Improved |
| Add the ℓq penalty to CE on all training data (ablation) | CE training · CIFAR-10, symmetric noise 0.4 | accuracy: 65.5 -> 71.6 | Improved |
| Apply CutMix to all training data (ablation) | CE training · CIFAR-10, symmetric noise 0.4 | accuracy: 65.5 -> 87.0 | Improved |
| Add the ℓq penalty to CE + SPR (ablation) | CE + SPR · CIFAR-10, symmetric noise 0.4 | accuracy: 80.4 -> 88.5 | Improved |
| Add CutMix to CE + SPR (ablation) | CE + SPR · CIFAR-10, symmetric noise 0.4 | accuracy: 80.4 -> 89.2 | Improved |
| Add CutMix to CE + SPR + ℓq (ablation) | Full SPR · CIFAR-10, symmetric noise 0.4 | accuracy: 88.5 -> 91.0 | Improved |
| Add the ℓq penalty to CE + SPR + CutMix (ablation) | Full SPR · CIFAR-10, symmetric noise 0.4 | accuracy: 89.2 -> 91.0 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DivideMix — DivideMix: Learning with Noisy Labels as Semi-supervised Learning (ICLR 2020) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Knockoffs-SPR: Clean Sample Selection in Learning With Noisy Labels (TPAMI 2023 · similarity 0.91)
- Safeguarded Dynamic Label Regression for Noisy Supervision (AAAI 2019 · similarity 0.90)
- FINE Samples for Learning with Noisy Labels (NeurIPS 2021 · similarity 0.90)
- Handling Label Noise via Instance-Level Difficulty Modeling and Dynamic Optimization (NeurIPS 2025 · similarity 0.89)
- Label Noise: Ignorance Is Bliss (NeurIPS 2024 · similarity 0.88)
- Learning with Open-world Noisy Data via Class-independent Margin in Dual Representation Space (AAAI 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.
