Research map / Optimization / Optimizers
Learning to Reweight Examples for Robust Deep Learning
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 |
|---|---|---|---|
| Use validation-guided online example reweighting with 1,000 clean examples instead of the strongest listed baseline, Random +FT | Ours · CIFAR-10 UniformFlip 40% noise | test accuracy: 86.55 $\pm$ 0.24 -> 86.92 $\pm$ 0.19 | Helped |
| Use validation-guided online example reweighting with 1,000 clean examples instead of MentorNet +FT | Ours · CIFAR-100 UniformFlip 40% noise | test accuracy: 59 -> 61.34 $\pm$ 2.06 | Helped |
| Use validation-guided online example reweighting with 10 clean images per class instead of the strongest listed baseline, Weighted +FT | Ours · CIFAR-10 BackgroundFlip 40% noise | test accuracy: 85.98 $\pm$ 0.47 -> 86.73 $\pm$ 0.48 | Helped |
| Use validation-guided online example reweighting with 10 clean images per class instead of S-Model +Conf +ES +FT | Ours · CIFAR-100 BackgroundFlip 40% noise | test accuracy: 55.75 $\pm$ 1.26 -> 59.30 $\pm$ 0.60 | Helped |
| Use random example weights instead of uniform training weights | CIFAR models under UniformFlip · CIFAR-10 UniformFlip 40% noise | test accuracy: 67.97 $\pm$ 0.62 -> 86.06 $\pm$ 0.32. | Helped |
| Use random example weights instead of uniform training weights | CIFAR models under UniformFlip · CIFAR-100 UniformFlip 40% noise | test accuracy: 50.66 $\pm$ 0.24 -> 58.01 $\pm$ 0.37 | Helped |
| Use random example weights instead of uniform training weights | CIFAR models under BackgroundFlip · CIFAR-10 BackgroundFlip 40% noise | test accuracy: 59.54 $\pm$ 2.16 -> 69.51 $\pm$ 1.36 | Helped |
| Use random example weights instead of uniform training weights | CIFAR models under BackgroundFlip · CIFAR-100 BackgroundFlip 40% noise | test accuracy: 37.82 $\pm$ 0.69 -> 36.56 $\pm$ 0.44 | Hurt |
| Add early stopping to S-Model (ablation) | S-Model on CIFAR-10 BackgroundFlip · CIFAR-10 BackgroundFlip 40% noise | test accuracy: + over 10% | Helped |
| Increase the clean validation set from 15 images to more than 100 images (ablation) | Online reweighting on CIFAR-10 BackgroundFlip · CIFAR-10 BackgroundFlip | test accuracy: null | No clear effect |
| Use only 15 validation images for online reweighting (ablation) | Online reweighting on CIFAR-10 BackgroundFlip · CIFAR-10 BackgroundFlip | test accuracy: 2% drop | Hurt |
| Increase label noise from 0% to 50% (ablation) | Ours on CIFAR-10 BackgroundFlip · CIFAR-10 BackgroundFlip | accuracy: 6% | Hurt |
Closest papers by meaning
- Learning to Reweight with Deep Interactions (AAAI 2021 · similarity 0.89)
- Meta-Weight-Net: Learning an Explicit Mapping For Sample Weighting (NeurIPS 2019 · similarity 0.89)
- Improving Generalization via Meta-Learning on Hard Samples (CVPR 2024 · similarity 0.87)
- Learning from Noisy Labels via Self-Taught On-the-Fly Meta Loss Rescaling (AAAI 2025 · similarity 0.84)
- Revive Re-weighting in Imbalanced Learning by Density Ratio Estimation (NeurIPS 2024 · similarity 0.84)
- Improving Accuracy-robustness Trade-off via Pixel Reweighted Adversarial Training (ICML 2024 · similarity 0.83)
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.