arXiv：cs.LG（机器学习，全量分类）· Laura Zichi, Gil Harari, Chuin Wei Tan, Marc L. Descoteaux, Albert Zhu, Menghang Wang, Yoel Zimmermann, H. T. Kung, Boris Kozinsky·· 14 小时前AI 评分33
BranchIP：为原子间势学习自适应等变计算
BranchIP: Learning Adaptive Equivariant Computation for Interatomic Potentials
AI 导读
BranchIP 是一个单模型框架，通过新的蒸馏损失学习自适应张量积计算，在异构催化与质子传导固体酸电解质两类体系上，将 MLIP 加速最高 2.4 倍、显存占用降低最高 2.6 倍，同时保持物理保真度。该自适应计算还能揭示哪些相互作用需要更深计算，以及计算深度与化学复杂度、动力学的关系，提供模型可解释性。
正文
Abstract:Equivariant machine learning interatomic potentials (MLIPs) have revolutionized atomistic modeling, but accurate treatment of complex materials and molecular systems demands expensive models. This limits simulation length- and time-scales, with tensor products a key computational bottleneck. The recent emergence of foundation-scale MLIPs further exacerbates this challenge. We present Branch Interatomic Potential (BranchIP), a single-model framework for learned adaptive tensor product computation, trained with a novel distillation loss. In our experiments on two systems of physical interest, a heterogeneous catalysis system and a proton-conducting solid acid electrolyte, BranchIP accelerates MLIPs across model sizes by up to $2.4\times$ while reducing memory usage by up to $2.6\times$. This is achieved while maintaining physical fidelity. Furthermore, the learned adaptive computation provides model interpretability by revealing which interactions demand deeper computation and showing how computational depth relates to chemical complexity and dynamics.
| Subjects: | Machine Learning (cs.LG); Applied Physics (physics.app-ph); Chemical Physics (physics.chem-ph); Computational Physics (physics.comp-ph) |
| Cite as: | arXiv:2610.02013 [cs.LG] |
| (or arXiv:2610.02013v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.02013 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Gil Harari [view email]
[v1]
Thu, 1 Oct 2026 16:39:30 UTC (10,601 KB)
来源：arXiv：cs.LG（机器学习，全量分类） · arxiv.org