Research map / Graph learning / GNN architectures
Transfer learning for atomistic simulations using GNNs and kernel mean embeddings
John Isak Texas Falk, Luigi Bonati, Pietro Novelli, Michele Parrinello, Massimiliano Pontil
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 |
|---|---|---|---|
| MEKRR-(SCN) instead of SchNet-FT | MEKRR · Fe/N2 D2 | RMSE: 2.0 -> 0.9 | Improved |
| MEKRR-(SchNet) instead of SchNet trained from scratch | SchNet · Cu/formate | RMSE: 6.0 -> 1.2 | Improved |
| MEKRR-(SCN) instead of GAP | GAP · Fe/N2 D4 | RMSE: 4.9 -> 2.7 | Improved |
| MEKRR-(SchNet) instead of SchNet-FT in transfer | SchNet-FT · D1->D2 | RMSE: 17.6 -> 8.0 | Improved |
| MEKRR-(SCN) instead of SchNet-FT in transfer | SchNet-FT · D3->D4 | RMSE: 116 -> 42 | Improved |
| MEKRR-(SchNet) instead of SchNet-FT | SchNet-FT · Fe/N2 D4 | RMSE: 3.2 -> 3.3 | No clear effect |
| Gaussian kernel (MEKRR) instead of linear kernel (ablation) | KRR alpha=1 · Fe/N2 D1 | RMSE: 21 -> 0.1 | Improved |
| Gaussian kernel (MEKRR) instead of linear kernel (ablation) | KRR alpha=0 · Fe/N2 D4 | RMSE: 146 -> 2.2 | Improved |
| species-weighted alpha=1 instead of shared weights alpha=0 (ablation) | MEKRR · Fe/N2 D1 | RMSE: 0.3 -> 0.1 | Improved |
| species-weighted alpha=1 instead of shared weights alpha=0 (ablation) | MEKRR · Fe/N2 D4 | RMSE: 2.2 -> 3.3 | Hurt |
| deeper representation layer 5 instead of layer 1 (ablation) | MEKRR alpha=1 · Fe/N2 D2 | RMSE: 1.7 -> 0.9 | Improved |
| deeper representation layer 5 instead of layer 1 (ablation) | MEKRR alpha=0 · Fe/N2 D2 | RMSE: 1.9 -> 1.4 | Improved |
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 0.88)
- ELoRA: Low-Rank Adaptation for Equivariant GNNs (ICML 2025 · similarity 0.88)
- Pushing the Limits of All-Atom Geometric Graph Neural Networks: Pre-Training, Scaling, and Zero-Shot Transfer (ICLR 2025 · similarity 0.88)
- Benchmarking Compositional Generalisation for Machine Learning Interatomic Potentials (NeurIPS 2026 · similarity 0.87)
- Learning Condensed Graph via Differentiable Atom Mapping for Reaction Yield Prediction (ICML 2025 · similarity 0.87)
- Stein Kernelized Molecular Dynamics for Active Learning of Interatomic Potentials (NeurIPS 2026 · 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.
