Research map / Learning theory and algorithms / Prediction and decision losses
Data Pruning via Moving-one-Sample-out
Haoru Tan, Sitong Wu, Fei Du, Yukang Chen, Zhibin Wang, Fan Wang, XIAOJUAN QI
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 |
|---|---|---|---|
| partition the training set into 10 subsets instead of 1 (ablation) | MoSo · CIFAR-100 pruning ratio 0.2 | Top-1 accuracy: 74.35 -> 75.81 | Improved |
| partition the training set into 5 subsets instead of 1 (ablation) | MoSo · CIFAR-100 pruning ratio 0.2 | Top-1 accuracy: 74.35 -> 75.76 | Improved |
| partition into 10 subsets instead of 5 (ablation) | MoSo · CIFAR-100 pruning ratio 0.2 | Top-1 accuracy: 75.76 -> 75.81 | No clear effect |
| train surrogate for 200 epochs instead of 50 (ablation) | MoSo · CIFAR-100 pruning ratio 0.2 | Top-1 accuracy: 75.76 -> 76.58 | No clear effect |
| train surrogate for 150 epochs instead of 100 (ablation) | MoSo · CIFAR-100 pruning ratio 0.2 | Top-1 accuracy: 76.41 -> 76.19 | No clear effect |
| remove training-dynamics awareness (use last-epoch gradient norm only) (ablation) | MoSo · CIFAR-100 across pruning rates | Top-1 accuracy | Hurt |
| lower the time step sampling rate (ablation) | MoSo · CIFAR-100 pruning ratio 0.4 | Top-1 accuracy | Hurt |
| use MoSo instead of SOTA baselines at high pruning ratios | MoSo · CIFAR-100 and Tiny-ImageNet | Top-1 accuracy | Improved |
| use EL2N instead of random selection at high pruning ratios | EL2N · CIFAR-100 | Top-1 accuracy | Hurt |
| MoSo versus Herding under 20% label noise | MoSo · CIFAR-100 and Tiny-ImageNet noisy labels | Top-1 accuracy: less than 1% | No clear effect |
Closest papers by meaning
- Exploring Learning Complexity for Efficient Downstream Dataset Pruning (ICLR 2025 · similarity 0.86)
- Lightweight Dataset Pruning without Full Training via Example Difficulty and Prediction Uncertainty (ICML 2025 · similarity 0.86)
- Distilling the Knowledge in Data Pruning (ICML 2025 · similarity 0.85)
- Sparsest Models Elude Pruning: An Exposé of Pruning’s Current Capabilities (ICML 2024 · similarity 0.85)
- $\mathbb{D}^2$ Pruning: Message Passing for Balancing Diversity & Difficulty in Data Pruning (ICLR 2024 · similarity 0.85)
- Data Mixture Optimization: A Multi-fidelity Multi-scale Bayesian Framework (NeurIPS 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.
