Research map / Label-efficient and robust learning / Noisy labels
Batch Loss Score for Dynamic Data Pruning
Qing Zhou, Bingxuan Zhao, Tao Yang, Hongyuan Zhang, Junyu Gao, Qi Wang
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 |
|---|---|---|---|
| add EMA smoothing to batch-loss score (ablation) | BLS-InfoBatch · CIFAR100 ResNet18 | accuracy (%): 77.9 -> 78.3 | Improved |
| add EMA smoothing to batch-loss score (ablation) | BLS-SeTa · CIFAR100 ResNet50 | accuracy (%): 79.9 -> 80.8 | Improved |
| add EMA smoothing to batch-loss score (ablation) | BLS-InfoBatch · CIFAR100 ResNet50 | accuracy (%): 79.9 -> 80.6 | Improved |
| replace per-sample loss with BLS in InfoBatch | InfoBatch · CIFAR100 ResNet18 30% pruning | accuracy (%): 78.2 -> 78.4 | No clear effect |
| replace per-sample loss with BLS in SeTa | SeTa · NoCaps Val Overall (ToCa) | CIDEr: 71.5 -> 71.2 | No clear effect |
| replace per-sample loss with BLS in InfoBatch | InfoBatch · NoCaps Val Overall (ToCa) | CIDEr: 70.2 -> 70.6 | No clear effect |
| replace per-sample loss with BLS in InfoBatch at 50% pruning | InfoBatch · CIFAR100 ResNet18 50% pruning | accuracy (%): 78.1 -> 77.7 | No clear effect |
| prune 22.9% of ImageNet-1K with BLS-InfoBatch | BLS-InfoBatch · ImageNet-1K Vim | accuracy (%): 75.7 -> 75.6 | No clear effect |
| BLS-SeTa pruning on YOLOv5n segmentation | BLS-SeTa · COCO instance segmentation | mAP.5: 29.5 -> 31.7 | Improved |
| BLS-InfoBatch pruning on CIFAR100 classification with YOLOv5n | BLS-InfoBatch · CIFAR100 YOLOv5n | Acc@1: 58.0 -> 57.6 | Hurt |
| raise EMA decay alpha from 0.5 to 0.7-0.8 (ablation) | BLS · CIFAR100 ResNet18 | accuracy (%): 79.0-79.1 -> 79.4 | Improved |
| set EMA decay alpha to 1.0 (ablation) | BLS · CIFAR100 ResNet18/50 | pruning ratio | Hurt |
Closest papers by meaning
- InfoBatch: Lossless Training Speed Up by Unbiased Dynamic Data Pruning (ICLR 2024 · similarity 0.87)
- Lightweight Dataset Pruning without Full Training via Example Difficulty and Prediction Uncertainty (ICML 2025 · similarity 0.87)
- Exploring Learning Complexity for Efficient Downstream Dataset Pruning (ICLR 2025 · similarity 0.86)
- Scale Efficient Training for Large Datasets (CVPR 2025 · similarity 0.86)
- DRoP: Distributionally Robust Data Pruning (ICLR 2025 · similarity 0.86)
- Batch Normalization Is Blind to the First and Second Derivatives of the Loss (AAAI 2024 · 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.
