Research map / Label-efficient and robust learning / Noisy labels
Learning With Noisy Labels via Sparse Regularization
Xiong Zhou, Xianming Liu, Chenyang Wang, Deming Zhai, Junjun Jiang, Xiangyang Ji
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 sparse regularization (SR) to CE | CE · MNIST (0.8 symmetric noise) | test accuracy: 22.66 -> 98.06 | Improved |
| add sparse regularization (SR) to CE | CE · CIFAR-10 (0.8 symmetric noise) | test accuracy: 19.55 -> 51.13 | Improved |
| add sparse regularization (SR) to CE | CE · CIFAR-100 (0.8 symmetric noise) | test accuracy: 7.59 -> 17.35 | Improved |
| add sparse regularization (SR) to CE | CE · CIFAR-100 (clean) | test accuracy: 71.33 -> 72.19 | Improved |
| add SR to FL | FL · WebVision (Mini, ResNet-50, top-1 val) | top-1 accuracy: 63.80 -> 70.28 | Improved |
| add SR to CE | CE · WebVision (Mini, ResNet-50, top-1 val) | top-1 accuracy: 66.96 -> 69.12 | Improved |
| add SR to CE | CE · CIFAR-10 (0.4 asymmetric noise) | test accuracy: 74.67 -> 79.29 | Improved |
| add SR to CE for imbalanced classification | CE · CIFAR-10 (long-tailed 0.01) | test accuracy: 64.16 -> 69.78 | Improved |
| add SR to CE for imbalanced classification | CE · CIFAR-100 (long-tailed 0.01) | test accuracy: 35.17 -> 41.24 | Improved |
| use GCE or APL instead of CE+SR under heavy noise | CE+SR · CIFAR-100 (0.8 symmetric noise) | test accuracy: 17.35 -> 24.91 | Hurt |
| use GCE instead of CE on clean data | CE · CIFAR-100 (clean) | test accuracy: 71.33 -> 63.09 | Hurt |
| increase lambda of l_p regularization to 7 or more (with tau=1, p=0.1) (ablation) | CE+SR · CIFAR-10 (0.6 symmetric noise) | test accuracy | Mixed |
Closest papers by meaning
- $\epsilon$-Softmax: Approximating One-Hot Vectors for Mitigating Label Noise (NeurIPS 2024 · similarity 0.92)
- Robust Training under Label Noise by Over-parameterization (ICML 2022 · similarity 0.92)
- Coresets for Robust Training of Deep Neural Networks against Noisy Labels (NeurIPS 2020 · similarity 0.91)
- Normalized Loss Functions for Deep Learning with Noisy Labels (ICML 2020 · similarity 0.91)
- On Learning Contrastive Representations for Learning With Noisy Labels (CVPR 2022 · similarity 0.91)
- Enhancing Noise-Robust Losses for Large-Scale Noisy Data Learning (AAAI 2025 · 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.
