Research map / Architectures and efficiency / Pruning and efficient tuning
Inconsistency Biases in Dynamic Data Pruning
Qing Zhou, Tao Yang, Bingxuan Zhao, 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 resampling to LWP and CTR (ablation) | RePB · CIFAR100 ResNet18 | accuracy: 19.0 -> 78.4 | Improved |
| add local window pruning (LWP) to resampling (ablation) | RePB · CIFAR100 ResNet18 | accuracy: 77.5 -> 78.1 | Improved |
| add cumulative temporal rescaling (CTR) to resampling (ablation) | RePB · CIFAR100 ResNet18 | accuracy: 77.5 -> 78.2 | Improved |
| add CTR to resampling plus LWP (ablation) | RePB · CIFAR100 ResNet18 | accuracy: 78.1 -> 78.4 | Improved |
| add LWP and CTR to resampling (ablation) | RePB · CIFAR100 ResNet50 | accuracy: 79.9 -> 80.8 | Improved |
| use RePB instead of InfoBatch at same epochs, 50% pruning | RePB · CIFAR100 ResNet18 | accuracy: 77.7 -> 78.1 | Improved |
| use RePB instead of InfoBatch, 70% pruning | RePB · CIFAR10 ResNet18 | accuracy: 94.7 -> 94.9 | Improved |
| integrate Salaun et al. sampling into RePB windows (ablation) | RePB · CIFAR100 ResNet18 | accuracy: 77.7 -> 78.2 | Improved |
| use gradient norm instead of loss as importance score (ablation) | RePB · CIFAR100 ResNet18 30% | accuracy: 78.4 -> 78.6 | No clear effect |
| use RePB instead of InfoBatch on ToCa captioning | RePB · ToCa NoCaps | CIDEr: 69.2 -> 70.5 | Improved |
| use RePB instead of InfoBatch on scene text recognition | RePB · MJ+ST IIIT5k | accuracy: 95.9 -> 96.1 | Improved |
| apply RePB to multi-view stereo | RePB · WHU-MVS | MAE (m): 0.1185 -> 0.1147 | Improved |
Closest papers by meaning
- Spanning Training Progress: Temporal Dual-Depth Scoring (TDDS) for Enhanced Dataset Pruning (CVPR 2024 · similarity 0.87)
- Lightweight Dataset Pruning without Full Training via Example Difficulty and Prediction Uncertainty (ICML 2025 · similarity 0.86)
- DRoP: Distributionally Robust Data Pruning (ICLR 2025 · similarity 0.85)
- InfoBatch: Lossless Training Speed Up by Unbiased Dynamic Data Pruning (ICLR 2024 · similarity 0.85)
- Exploring Learning Complexity for Efficient Downstream Dataset Pruning (ICLR 2025 · similarity 0.85)
- GDeR: Safeguarding Efficiency, Balancing, and Robustness via Prototypical Graph Pruning (NeurIPS 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.
