Research map / Label-efficient and robust learning / Noisy labels
Learning Fast Sample Re-Weighting Without Reward Data
Zizhao Zhang, Tomas Pfister
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 partial layers (feature sharing) for the meta model (ablation) | FSR · CIFAR10 40% uniform noise, ResNet32 | accuracy | Improved |
| use partial meta layers (feature sharing) instead of full model | FSR vs L2R · training cost | training time overhead: 47% ~ 87% vs 319% ~ 443% | Improved |
| use partial meta layers instead of full model | FSR vs L2R · GPU memory | memory overhead: 1.9% ~ 34% vs 54% ~ 270% | Improved |
| add MixUp for the weighted loss (ablation) | FSR · noise robust training | dictionary purity and accuracy | Improved |
| remove MixUp from forgetting-event pusher (ablation) | FSR (forget-event pusher) · CIFAR100 20%/40% noise | accuracy: 68.1 / 60.9 | Improved |
| add momentum re-labeling (ablation) | FSR · CIFAR10 20% uniform noise | accuracy, purity | Improved |
| use meta-margin instead of negative-loss or AUM pusher (ablation) | FSR · CIFAR100 20%/40% noise | accuracy | Improved |
| use forgetting-event inverse as pusher instead of meta-margin (ablation) | FSR · CIFAR100 noise | accuracy | Hurt |
| add MixUp or label smoothing for long-tailed recognition (ablation) | FSR · CIFAR long-tailed | accuracy | No clear effect |
| increase real reward data size (ablation) | L2R · CIFAR10 40% uniform noise | accuracy | No clear effect |
| vary dictionary size (ablation) | FSR · CIFAR10 40% uniform noise | accuracy | No clear effect |
| sweep hyper-parameters (ablation) | FSR · CIFAR100 40% uniform noise | accuracy: ~2% | No clear effect |
Closest papers by meaning
- Dirichlet-based Per-Sample Weighting by Transition Matrix for Noisy Label Learning (ICLR 2024 · similarity 0.86)
- Group-robust Sample Reweighting for Subpopulation Shifts via Influence Functions (ICLR 2025 · similarity 0.85)
- Optimal Transport Reweighting for Robust Learning under Spurious Correlations and Label Noise (NeurIPS 2026 · similarity 0.85)
- Improving Generalization via Meta-Learning on Hard Samples (CVPR 2024 · similarity 0.85)
- Sample-Optimal Agnostic Boosting with Unlabeled Data (ICML 2025 · similarity 0.85)
- Latent Score-Based Reweighting for Robust Classification on Imbalanced Tabular Data (ICML 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.
