Research map / Label-efficient and robust learning / Noisy labels
Meta-Weight-Net: Learning an Explicit Mapping For Sample Weighting
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 |
|---|---|---|---|
| Learn sample weights with Meta-Weight-Net using unbiased meta-data instead of a fixed weighting scheme | ResNet-32 · Long-Tailed CIFAR-10, imbalanced settings 200 through 10 | accuracy: 68.89 -> 68.91; 74.57 -> 75.21; 79.27 -> 80.06; 84.36 -> 84.94; 87.49 -> 87.84 | Helped |
| Learn sample weights with Meta-Weight-Net using unbiased meta-data instead of a fixed weighting scheme | ResNet-32 · Long-Tailed CIFAR-100, imbalanced settings 100 through 10 | accuracy: 41.83 -> 42.09; 46.40 -> 46.74; 52.59 -> 54.37; 57.99 -> 58.46 | Helped |
| Learn sample weights with Meta-Weight-Net instead of fine-tuning on meta-data | ResNet-32 · Long-Tailed CIFAR-10, imbalance 1 | accuracy: 93.23 -> 92.66 | Hurt |
| Learn sample weights with Meta-Weight-Net instead of fine-tuning on meta-data | ResNet-32 · Long-Tailed CIFAR-100, imbalance 200 | accuracy: 38.22 -> 37.91 | Hurt |
| Use Meta-Weight-Net rather than the strongest reported baseline | WRN-28-10 · CIFAR-10, 40% and 60% uniform noise | accuracy: 88.28$\pm$0.03 -> 89.27±0.28; 83.49$\pm$0.24 -> 84.07±0.33 | Helped |
| Use Meta-Weight-Net rather than the strongest reported baseline | WRN-28-10 · CIFAR-100, 40% and 60% uniform noise | accuracy: 61.39$\pm$3.99 -> 67.73$\pm$0.26; 50.81$\pm$1.00 -> 58.75$\pm$0.11 | Helped |
| Use Meta-Weight-Net rather than the strongest reported baseline | ResNet-32 · CIFAR-10, 20% flip noise | accuracy: 89.68 -> 90.33 | Helped |
| Use Meta-Weight-Net rather than the strongest reported baseline | ResNet-32 · CIFAR-10, 40% flip noise | accuracy: 88.92 -> 87.54 | Hurt |
| Use Meta-Weight-Net rather than the strongest reported baseline | ResNet-32 · CIFAR-100, 20% flip noise | accuracy: 63.07 -> 64.22 | Helped |
| Use Meta-Weight-Net rather than the strongest reported baseline | ResNet-32 · CIFAR-100, 40% flip noise | accuracy: 62.22 -> 58.64 | Hurt |
| Learn sample weights with Meta-Weight-Net rather than MLNT | ResNet-50 · Clothing1M test | accuracy: 73.47 -> 73.72 | Helped |
Closest papers by meaning
- Delving into Sample Loss Curve to Embrace Noisy and Imbalanced Data (AAAI 2022 · similarity 0.90)
- Learning to Reweight Examples for Robust Deep Learning (ICML 2018 · similarity 0.89)
- Improving Generalization via Meta-Learning on Hard Samples (CVPR 2024 · similarity 0.86)
- Revive Re-weighting in Imbalanced Learning by Density Ratio Estimation (NeurIPS 2024 · similarity 0.86)
- Optimizing importance weighting in the presence of sub-population shifts (ICLR 2025 · similarity 0.85)
- Thumb on the Scale: Optimal Loss Weighting in Last Layer Retraining (NeurIPS 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.