Training machine learning interatomic potentials that are both computationally and data-efficient is a key challenge for enabling their routine use in atomistic simulations.To this effect,we introduce franken,a scalab...Training machine learning interatomic potentials that are both computationally and data-efficient is a key challenge for enabling their routine use in atomistic simulations.To this effect,we introduce franken,a scalable and lightweight transfer learning framework that extracts atomic descriptors from pre-trained graph neural networks and transfers them to new systems using random Fourier features—an efficient and scalable approximation of kernel methods.It also provides a closed-form finetuning strategy for general-purpose potentials such as MACE-MP0,enabling fast and accurate adaptation to new systems or levels of quantum mechanical theory with minimal hyperparameter tuning.On a benchmark dataset of 27 transition metals,franken outperforms optimized kernelbased methods in both training time and accuracy,reducing model training from tens of hours to minutes on a single GPU.We further demonstrate the framework’s strong data-efficiency by training stable and accurate potentials for bulk water and the Pt(111)/water interface using just tens of training structures.Our open-source implementation(https://franken.readthedocs.io)offers a fast and practical solution for training potentials and deploying them for molecular dynamics simulations across diverse systems.展开更多
基金support of the Data Science and Computation Facility at the Fondazione Istituto Italiano di Tecnologia and the CINECA award under the ISCRA initiativeThis work was partially funded by the European Union—NextGenerationEU initiative and the Italian National Recovery and Resilience Plan (PNRR) from the Ministry of University and Research (MUR), under Project PE0000013 CUP J53C22003010006 “Future Artificial Intelligence Research (FAIR)”+3 种基金L.R. acknowledges the financial support of the European Research Council (grant SLING 819789)the European Commission (Horizon Europe grant ELIAS 101120237)the Ministry of Education, University and Research (FARE grant ML4IP R205T7J2KPgrant BAC FAIR PE00000013 CUP J33C24000420007 funded by the EU—NGEU).
文摘Training machine learning interatomic potentials that are both computationally and data-efficient is a key challenge for enabling their routine use in atomistic simulations.To this effect,we introduce franken,a scalable and lightweight transfer learning framework that extracts atomic descriptors from pre-trained graph neural networks and transfers them to new systems using random Fourier features—an efficient and scalable approximation of kernel methods.It also provides a closed-form finetuning strategy for general-purpose potentials such as MACE-MP0,enabling fast and accurate adaptation to new systems or levels of quantum mechanical theory with minimal hyperparameter tuning.On a benchmark dataset of 27 transition metals,franken outperforms optimized kernelbased methods in both training time and accuracy,reducing model training from tens of hours to minutes on a single GPU.We further demonstrate the framework’s strong data-efficiency by training stable and accurate potentials for bulk water and the Pt(111)/water interface using just tens of training structures.Our open-source implementation(https://franken.readthedocs.io)offers a fast and practical solution for training potentials and deploying them for molecular dynamics simulations across diverse systems.
