Research map / Label-efficient and robust learning / Noisy labels
Delving into Sample Loss Curve to Embrace Noisy and Imbalanced Data
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 loss curves and CurveNet to assign sample weights for noisy and imbalanced data | ResNet-32 classifier · CIFAR10, varying noise and imbalance | mean accuracy: 74.13 -> 75.70 | Helped |
| Use loss curves and CurveNet to assign sample weights for noisy and imbalanced data | ResNet-32 classifier · CIFAR100, varying noise and imbalance | mean accuracy: 49.28 -> 50.49 | Helped |
| Use the proposed weighting method instead of MW-Net(noise) | CIFAR10, imbalance ratio 1, noise rate 0.2 · CIFAR10, noise rate 0.2 | accuracy: 90.80 -> 92.01 | Helped |
| Use the proposed weighting method instead of MW-Net(imb) | CIFAR100, imbalance ratio 20, noise rate 0.4 · CIFAR100, noise rate 0.4 | accuracy: 34.90 -> 37.88 | Helped |
| Use the proposed method instead of MW-Net | Clothing1M · Clothing1M | accuracy: 73.72 -> 74.41 | Helped |
| Use the proposed method instead of Distill | Food-101N · Food-101N | accuracy: 87.57 -> 87.19 | Hurt |
| Add DRW to the proposed method (ablation) | WRN-16-8 · CIFAR10, imbalance ratio 10, noise rate 0.2 | accuracy: 85.94 -> 86.52 | Helped |
| Add DRW to the proposed method (ablation) | WRN-16-8 · CIFAR10, imbalance ratio 20, noise rate 0.6 | accuracy: 65.77 -> 66.78 | Helped |
| Reduce the unbiased meta set from 100 to 10 samples (ablation) | Proposed method · CIFAR10, noise rate 0.2, imbalance ratio 20 | accuracy: 81.61 -> 79.13 | Hurt |
| Select low-loss training samples instead of using extra unbiased meta data (ablation) | Proposed method · CIFAR10, noise rate 0.2, imbalance ratio 20 | accuracy: 78.81 -> 79.02 | Helped |
| Skip 3 classifier layers during meta optimization (ablation) | Proposed method · CIFAR10, varying noise and imbalance | training time: 239.94 -> 41.98 | Helped |
| Skip 3 classifier layers during meta optimization (ablation) | Proposed method · CIFAR10, varying noise and imbalance | mean accuracy: 82.46 -> 81.53 | Hurt |
Closest papers by meaning
- Meta-Weight-Net: Learning an Explicit Mapping For Sample Weighting (NeurIPS 2019 · similarity 0.90)
- Mitigating Dataset Bias by Using Per-Sample Gradient (ICLR 2023 · similarity 0.86)
- Learning from Noisy Labels via Self-Taught On-the-Fly Meta Loss Rescaling (AAAI 2025 · similarity 0.86)
- A Simple Remedy for Dataset Bias via Self-Influence: A Mislabeled Sample Perspective (NeurIPS 2024 · similarity 0.86)
- Improving Generalization via Meta-Learning on Hard Samples (CVPR 2024 · similarity 0.85)
- Enhancing Noise-Robust Losses for Large-Scale Noisy Data Learning (AAAI 2025 · similarity 0.84)
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.