跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2611.21116 · v1

Out-of-training properties in frozen interatomic potential embeddings are decodable and ordered by physical scale

首次提交 2026-11-16 21:59
版本更新 2026-11-16 21:59
雷达观测 2026-11-17 17:11
arXiv ↗PDF ↗

摘要

Machine-learned interatomic potentials (MLIPs) are trained on energies, forces and stresses, yet their frozen representations can predict other materials properties. We probe ORB-v3 and UMA-S across layers, read-outs and equivariant channels for nine properties of 154,875 Materials Project crystals. Both models outperform composition-only descriptors on all targets, including five outside their training objectives; errors in physical units help assess potential screening applications. Optimal depth varies by property, consistent with a physical-scale ordering. Random-weight controls reveal strongly property-dependent benefits of pretraining. UMA-S's rotation-invariant angular channels improve crystal-system classification, while higher scores from its full tensor disappear under rotation; matched-model controls do not isolate equivariance as the explanation. Accuracy varies across different probes, but nearest-neighbour prediction often approaches XGBoost performance, even though materials do not form distinct clusters according to their properties. We also test whether a linear predictor trained on one model's embeddings can predict properties using the other model's embeddings after a learned linear transformation. Predictive accuracy is largely preserved, despite differences in which atoms the two models place close together. These results guide the reuse of frozen MLIP representations for materials prediction.

作者备注:61 pages (9 main text), 18 figures, 23 tables. Accepted at AI4Mat and AI4Science workshops, NeurIPS 2026. Code: https://github.com/Natalija-Stepurko/mlip-embedding-probes ; ORB-v3 embeddings: https://huggingface.co/datasets/NatalijaStepurko/orb-mp-embeddings ; UMA-S embeddings (gated): https://huggingface.co/datasets/NatalijaStepurko/uma-mp-embeddings-gated

AI 论文解读

基于摘要 · 非全文审阅

此版本尚未生成 AI 解读。解读由后台任务生成并按论文版本保存,登录后共享已有结果。

版本差异

目前只收录了一个版本。收录下一版后,可以对照标题、作者、分类和摘要。

观测到的版本

相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。

v1 · 2026-11-16 21:59 · 最新观测版本

Out-of-training properties in frozen interatomic potential embeddings are decodable and ordered by physical scale

Machine-learned interatomic potentials (MLIPs) are trained on energies, forces and stresses, yet their frozen representations can predict other materials properties. We probe ORB-v3 and UMA-S across layers, read-outs and equivariant channels for nine properties of 154,875 Materials Project crystals. Both models outperform composition-only descriptors on all targets, including five outside their training objectives; errors in physical units help assess potential screening applications. Optimal depth varies by property, consistent with a physical-scale ordering. Random-weight controls reveal strongly property-dependent benefits of pretraining. UMA-S's rotation-invariant angular channels improve crystal-system classification, while higher scores from its full tensor disappear under rotation; matched-model controls do not isolate equivariance as the explanation. Accuracy varies across different probes, but nearest-neighbour prediction often approaches XGBoost performance, even though materials do not form distinct clusters according to their properties. We also test whether a linear predictor trained on one model's embeddings can predict properties using the other model's embeddings after a learned linear transformation. Predictive accuracy is largely preserved, despite differences in which atoms the two models place close together. These results guide the reuse of frozen MLIP representations for materials prediction.

查看此版本解读 →查看此版本原文 ↗