Research map / Architectures and efficiency / Pruning and efficient tuning
PFT: Phonon Fine-tuning for Machine Learned Interatomic Potentials
Teddy Koker, Abhijeet Sadashiv Gangan, Mit Kotak, Jaime Marian, Tess Smidt
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 |
|---|---|---|---|
| fine-tune with Hessian (force constant) loss PFT | Nequix MP · MDR Phonon test | average MAE across phonon properties: 55% average reduction; max frequency 24 -> 12 K | Improved |
| apply PFT to a stronger base trained on OMat24 data (Nequix OAM) | Nequix OAM · MDR Phonon test | average MAE across phonon properties: 50% average relative improvement; max frequency 17 -> 9 K | Improved |
| fine-tune energy/force/stress directly on phonon displacement data instead of Hessian loss (ablation) | Nequix MP · MDR Phonon test | max phonon frequency MAE (K): 24 -> 182 | Hurt |
| add co-training with upstream MPtrj data (ablation) | Nequix MP PFT · Matbench Discovery stability | F1: 0.301 -> 0.748 | Improved |
| add co-training with upstream MPtrj data (ablation) | Nequix MP PFT · MDR Phonon test | max phonon frequency MAE (K): 10 -> 12 | Hurt |
| add co-training with upstream MPtrj data (ablation) | Nequix MP PFT · Matbench Discovery geometry optimization | RMSD (A): 0.089 -> 0.087 | Improved |
| compute force constants analytically via autodiff instead of finite displacement (ablation) | Nequix MP PFT · MDR Phonon test | phonon property MAE: max frequency 12 -> 12 | No clear effect |
| apply PFT to MACE-MP-0 without co-training | MACE-MP-0 · MDR Phonon test | max phonon frequency MAE (K): 61 -> 19 | Improved |
| apply PFT (generalization to third-order force constants) | Nequix MP · third-order force constants | MAE of Phi(3) (meV/A^3): 10.52 -> 8.35 | Improved |
| apply PFT to improve thermal conductivity | Nequix MP · Matbench Discovery thermal conductivity | kappa SRME: 0.446 -> 0.307 | Improved |
| apply PFT to improve thermal conductivity | Nequix OAM · Matbench Discovery thermal conductivity | kappa SRME: 0.267 -> 0.198 | Improved |
| apply PFT with co-training | Nequix MP · Matbench Discovery stability | F1: 0.751 -> 0.748 | No clear effect |
Closest papers by meaning
- Learning Smooth and Expressive Interatomic Potentials for Physical Property Prediction (ICML 2025 · similarity 0.90)
- Learning the Electronic Hamiltonian of Large Atomic Structures (ICML 2025 · similarity 0.86)
- Enhancing the Scalability and Applicability of Kohn-Sham Hamiltonians for Molecular Systems (ICLR 2025 · similarity 0.86)
- Physics-Informed Weakly Supervised Learning For Interatomic Potentials (ICML 2025 · similarity 0.86)
- ELoRA: Low-Rank Adaptation for Equivariant GNNs (ICML 2025 · similarity 0.85)
- The dark side of the forces: assessing non-conservative force models for atomistic machine learning (ICML 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.
