Research map / Label-efficient and robust learning / Noisy labels
Early-Learning Regularization Prevents Memorization of Noisy Labels
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 |
|---|---|---|---|
| Apply ELR instead of SL | ResNet34 · CIFAR-10, symmetric 80% noise | accuracy: 68.12 $\pm$ 0.81 -> 73.86 $\pm$ 0.61 | Helped |
| Apply ELR instead of SL | ResNet34 · CIFAR-100, symmetric 20% noise | accuracy: 70.38 $\pm$ 0.13 -> 74.21 $\pm$ 0.22 | Helped |
| Use cosine annealing learning rate with ELR+ | ELR+ · CIFAR-10, symmetric 80% noise | accuracy: 91.1 -> 93.3 | Helped |
| Use cosine annealing learning rate with ELR+ | ELR+ · CIFAR-100, symmetric 80% noise | accuracy: 58.2 -> 60.8 | Helped |
| Use ELR+ rather than DivideMix | CIFAR-100 · CIFAR-100, symmetric 50% noise | accuracy: 74.6 -> 72.4 | Hurt |
| Use ELR+ rather than DivideMix | WebVision training · ILSVRC12, top1 | top1 accuracy: 75.20 -> 70.29 | Hurt |
| Add weight averaging (ablation) | ELR+, one network with mixup · CIFAR-10, symmetric 40% noise | accuracy: 91.05 $\pm$ 0.13 -> 93.04 $\pm$ 0.12 | Helped |
| Add weight averaging (ablation) | ELR+, two networks with mixup · CIFAR-10, symmetric 80% noise | accuracy: 84.75 $\pm$ 0.26 -> 88.62 $\pm$ 0.26 | Helped |
| Add mixup (ablation) | ELR+, one network with weight averaging · CIFAR-10, symmetric 80% noise | accuracy: 76.50 $\pm$ 0.65 -> 87.23 $\pm$ 0.30 | Helped |
| Add mixup (ablation) | ELR+, two networks with weight averaging · CIFAR-10, symmetric 40% noise | accuracy: 92.95 $\pm$ 0.05 -> 93.68 $\pm$ 0.51 | Helped |
| Use two networks instead of one (ablation) | ELR+, with mixup and weight averaging · CIFAR-10, symmetric 80% noise | accuracy: 87.23 $\pm$ 0.30 -> 88.62 $\pm$ 0.26 | Helped |
| Use two networks instead of one (ablation) | ELR+, without mixup or weight averaging · CIFAR-10, symmetric 80% noise | accuracy: 72.54 $\pm$ 0.35 -> 73.49 $\pm$ 0.47 | Helped |
Closest papers by meaning
- On the Role of Label Noise in the Feature Learning Process (ICML 2025 · similarity 0.87)
- Pre-Trained Vision-Language Models as Noisy Partial Annotators (AAAI 2025 · similarity 0.87)
- Unveiling Memorization–Generalization Coexistence: A Case Study on Arithmetic Tasks with Label Noise (NeurIPS 2026 · similarity 0.87)
- Handling Label Noise via Instance-Level Difficulty Modeling and Dynamic Optimization (NeurIPS 2025 · similarity 0.86)
- Learning with Structural Labels for Learning with Noisy Labels (CVPR 2024 · similarity 0.86)
- To Grok or not to Grok: Disentangling Generalization and Memorization on Corrupted Algorithmic Datasets (ICLR 2024 · similarity 0.86)
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.