Research map / Probabilistic and scientific modeling / Molecules and proteins
FUSION: Dataset Pruning via Fusing Uncertainty with Structural Information for Optimal Neural Training in Crystal Property Prediction
Xiean Wang, Pin Chen, Liqin Tan, Yutong Lu, Qingsong Zou
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 FUSION (uncertainty + structure) instead of random pruning | ALIGNN · Perovskites OOD, 80% pruned | MAE: 0.223 -> 0.127 | Improved |
| use FUSION instead of random pruning | ALIGNN · Dielectric OOD, 80% pruned | MAE: 0.421 -> 0.319 | Improved |
| use FUSION instead of random pruning across 11 architectures | 11 models · Perovskites (ID) | MAE improvement: 13.65% average | Improved |
| use FUSION instead of random pruning across 11 architectures | 11 models · Dielectric (ID) | MAE improvement: 1.91% average | Improved |
| use FUSION instead of diversity sampling | ALIGNN · JDFT2D ID, 80% pruned | MAE: 12.7% MAE reduction | Improved |
| add structural weighting factor to uncertainty (full FUSION vs uncertainty only) (ablation) | FUSION · Dielectric ID, 80% pruned | MAE: 0.940 vs FUSION (161% worse) | Improved |
| add uncertainty to structural weighting factor (full FUSION vs weighting only) (ablation) | FUSION · Perovskites OOD, 80% pruned | MAE: 0.200 -> 0.127 | Improved |
| move from ID to OOD evaluation | FUSION vs second-best · JDFT2D 80% pruned | MAE gap: 9.752 -> 0.185 MAE gap | Hurt |
| prune a small optimal fraction (0-8.5%) of training data | various models · Matbench datasets | MAE: up to 12.67% MAE reduction | Improved |
| use ensemble-disagreement active learning for pruning | ALIGNN · JDFT2D ID, 80% pruned | MAE: 131.210 | Hurt |
| FUSION vs random pruning at 40% pruning | ALIGNN · Perovskites OOD, 40% pruned | MAE: 0.098 -> 0.100 | No clear effect |
Closest papers by meaning
- A Denoising Pre-training Framework for Accelerating Novel Material Discovery (AAAI 2025 · similarity 0.89)
- All that structure matches does not glitter (NeurIPS 2025 · similarity 0.89)
- A Diffusion-Based Pre-training Framework for Crystal Property Prediction (AAAI 2024 · similarity 0.88)
- Crystalformer: Infinitely Connected Attention for Periodic Structure Encoding (ICLR 2024 · similarity 0.86)
- Vector Field Oriented Diffusion Model for Crystal Material Generation (AAAI 2024 · similarity 0.86)
- Scalable Diffusion for Materials Generation (ICLR 2024 · similarity 0.86)
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.
