Research map / Label-efficient and robust learning / Noisy labels
Noise Attention Learning: Enhancing Noise Robustness by Gradient Scaling
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 target estimation (momentum label correction) (ablation) | NAL · CIFAR-10 symm 40% | accuracy (%): 89.47 -> 93.49 | Helped |
| add target estimation (ablation) | NAL · CIFAR-10 symm 80% | accuracy (%): 76.91 -> 80.98 | Helped |
| add target estimation (ablation) | NAL · CIFAR-100 asymm 40% | accuracy (%): 55.68 -> 74.73 | Helped |
| use learned attention branch instead of max softmax confidence as sample weight (ablation) | NAL · CIFAR-10 symm 40% | accuracy (%): 90.94 -> 93.49 | Helped |
| use learned attention branch instead of max softmax confidence as sample weight (ablation) | NAL · CIFAR-10 symm 80% | accuracy (%): fails to converge -> 80.98 | Helped |
| add reverse term RNAL (SCE-style) to NAL (ablation) | NAL · CIFAR-10 asymm 40% | accuracy (%): 92.09 -> 88.17 | Hurt |
| add reverse term RNAL (SCE-style) to NAL (ablation) | NAL · CIFAR-10 symm 80% | accuracy (%): 80.98 -> 80.20 | Hurt |
| add reverse term RNAL (SCE-style) to NAL (ablation) | NAL · CIFAR-100 symm 80% | accuracy (%): 36.77 -> 38.24 | Helped |
| NAL versus cross entropy training | NAL · CIFAR-10 symm 80% | accuracy (%): 53.82 -> 80.98 | Helped |
| NAL versus cross entropy training | NAL · CIFAR-100 asymm 40% | accuracy (%): 42.74 -> 74.73 | Helped |
| NAL versus best baseline at highest noise on CIFAR-100 symm 80% | NAL · CIFAR-100 symm 80% | accuracy (%): SAT 38.72 vs NAL 36.77 | Mixed |
| NAL trained from scratch versus ImageNet-pretrained baselines | NAL · Clothing1M | accuracy (%): 73.39 -> 73.58 | Helped |
Closest papers by meaning
- Handling Label Noise via Instance-Level Difficulty Modeling and Dynamic Optimization (NeurIPS 2025 · similarity 0.88)
- On the Role of Label Noise in the Feature Learning Process (ICML 2025 · similarity 0.88)
- How Does Label Noise Gradient Descent Improve Generalization in the Low SNR Regime? (NeurIPS 2025 · similarity 0.87)
- Mitigating Label Noise through Data Ambiguation (AAAI 2024 · similarity 0.87)
- Understanding and Mitigating the Label Noise in Pre-training on Downstream Tasks (ICLR 2024 · similarity 0.86)
- Enhancing Noise-Robust Losses for Large-Scale Noisy Data Learning (AAAI 2025 · 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.