Research map / Learning theory and algorithms / Prediction and decision losses
Hierarchical Dataset Selection for High-Quality Data Sharing
Xiaona Zhou, Yingyan Zeng, Ran Jin, Ismini Lourentzou
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 hierarchical group-level modeling (DaSH vs flat DaS) (ablation) | DaS (flat) · Digit-Five USPS | exploration steps: 163 -> 140 | Improved |
| add hierarchical group-level modeling (DaSH vs flat DaS) (ablation) | DaS (flat) · Digit-Five USPS | accuracy: 90.9 -> 91.2 | No clear effect |
| add hierarchy under 15-step exploration budget (ablation) | DaS (flat) · Digit-Five | accuracy: +8.8% on MNIST, +1.8% on USPS, +9.8% on MNIST-M, +4.5% on SYN | Improved |
| use mixed (imperfect) groups instead of perfect groups (ablation) | DaSH · Digit-Five and DomainNet | accuracy: within 1-2% in most domains | No clear effect |
| use cross-domain grouping instead of same-domain groups (ablation) | DaSH · Digit-Five USPS | accuracy: 91.2 -> 92.2 | No clear effect |
| use DaSH instead of FreeSel | FreeSel · Digit-Five | accuracy: 52.5 -> 78.3 | Improved |
| use DaSH instead of ActiveFT | ActiveFT · Digit-Five | accuracy: 52.1 -> 78.3 | Improved |
| use DaSH instead of Core-sets | Core-sets · DomainNet | accuracy: 70.2 -> 81.0 | Improved |
| use DaSH instead of FreeSel | FreeSel · DomainNet | accuracy: 77.7 -> 81.0 | Improved |
| apply DaSH selection to weak initial model (10% data) | Local model · Digit-Five MNIST | accuracy: 17.6 -> 31.5 | Improved |
| apply DaSH selection to weak initial model (20% data) | Local model · Digit-Five SVHN | accuracy: 21.2 -> 21.5 | No clear effect |
Closest papers by meaning
- Diversified Batch Selection for Training Acceleration (ICML 2024 · similarity 0.87)
- DsDm: Model-Aware Dataset Selection with Datamodels (ICML 2024 · similarity 0.87)
- Data-Efficient Learning via Clustering-Based Sensitivity Sampling: Foundation Models and Beyond (ICML 2024 · similarity 0.85)
- Most Influential Subset Selection: Challenges, Promises, and Beyond (NeurIPS 2024 · similarity 0.85)
- Harnessing Diversity for Important Data Selection in Pretraining Large Language Models (ICLR 2025 · similarity 0.84)
- SAVA: Scalable Learning-Agnostic Data Valuation (ICLR 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.
