Research map / Label-efficient and robust learning / Recognition and long-tail
Injecting Domain Knowledge from Empirical Interatomic Potentials to Neural Networks for Predicting Material Properties
Zeren Shui, Daniel S. Karls, Mingjian Wen, ilia Andreyevich Nikiforov, Ellad Tadmor, George Karypis
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 label augmentation with EIP labels (ablation) | SchNet · ANI-Al | configuration-level MAE (eV): 0.1693 -> 0.0845 | Improved |
| add label augmentation (ablation) | SOAPNet · KIM-Si | configuration-level MAE (eV): 0.7706 -> 0.5595 | Improved |
| add label augmentation (ablation) | CGCNN · AgAu | configuration-level MAE (eV): 1.6683 -> 1.6065 | Improved |
| add multi-task pretraining on EIP energies (ablation) | SchNet · AgAu | configuration-level MAE (eV): 0.7290 -> 0.3353 | Improved |
| add multi-task pretraining (ablation) | GemNet · KIM-Si | configuration-level MAE (eV): 0.5138 -> 0.5024 | Improved |
| combine multi-task pretraining and label augmentation (ablation) | SchNet · KIM-Si | configuration-level MAE (eV): 0.4805 -> 0.3719 | Improved |
| combine both strategies vs multi-task pretraining alone (ablation) | GemNet · AgAu | configuration-level MAE (eV): 0.5057 -> 0.6074 | Hurt |
| combine both strategies vs multi-task pretraining alone (ablation) | SchNet · AgAu | configuration-level MAE (eV): 0.3353 -> 0.3496 | Hurt |
| replace Tukey loss with MSE loss (ablation) | SchNet +LA · ANI-Al | configuration-level MAE (eV): 0.0845 -> 0.1031 | Hurt |
| replace Tukey loss with MSE loss (ablation) | SOAPNet +LA · KIM-Si | configuration-level MAE (eV): 0.5595 -> 0.5556 | No clear effect |
| label augmentation with only the worst EIP (ablation) | SOAPNet · KIM-Si | configuration-level MAE (eV): 0.7582 -> 0.766 | Hurt |
| label augmentation with only the best EIP vs default 8 EIPs (ablation) | SOAPNet · KIM-Si | configuration-level MAE (eV): 0.5673 -> 0.6084 | Hurt |
Closest papers by meaning
- The Importance of Being Scalable: Improving the Speed and Accuracy of Neural Network Interatomic Potentials Across Chemical Domains (NeurIPS 2024 · similarity 0.91)
- Learning the Electronic Hamiltonian of Large Atomic Structures (ICML 2025 · similarity 0.91)
- Learning Smooth and Expressive Interatomic Potentials for Physical Property Prediction (ICML 2025 · similarity 0.91)
- Physics-Informed Weakly Supervised Learning For Interatomic Potentials (ICML 2025 · similarity 0.90)
- Benchmarking Compositional Generalisation for Machine Learning Interatomic Potentials (NeurIPS 2026 · similarity 0.90)
- ELoRA: Low-Rank Adaptation for Equivariant GNNs (ICML 2025 · similarity 0.88)
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.
