{"title":"Lightweight and high-precision materials property prediction using pre-trained Graph Neural Networks and its application to small dataset","authors":"Kento Nishio, Kiyou Shibata, Teruyasu Mizoguchi","doi":"10.35848/1882-0786/ad2a06","DOIUrl":null,"url":null,"abstract":"\n Large data sets are essential for building deep learning models. However, generating large datasets with higher theoretical levels and larger computational models remains difficult due to the high cost of first-principles calculation. Here, we propose a lightweight and highly accurate machine learning approach using pre-trained Graph Neural Networks (GNNs) for industrially important but difficult to scale models. The proposed method was applied to a small dataset of graphene surface systems containing surface defects, and achieved comparable accuracy with six orders of magnitude faster learning than when the GNN was trained from scratch.","PeriodicalId":8093,"journal":{"name":"Applied Physics Express","volume":null,"pages":null},"PeriodicalIF":2.3000,"publicationDate":"2024-02-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Applied Physics Express","FirstCategoryId":"101","ListUrlMain":"https://doi.org/10.35848/1882-0786/ad2a06","RegionNum":4,"RegionCategory":"物理与天体物理","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"PHYSICS, APPLIED","Score":null,"Total":0}
引用次数: 0
Abstract
Large data sets are essential for building deep learning models. However, generating large datasets with higher theoretical levels and larger computational models remains difficult due to the high cost of first-principles calculation. Here, we propose a lightweight and highly accurate machine learning approach using pre-trained Graph Neural Networks (GNNs) for industrially important but difficult to scale models. The proposed method was applied to a small dataset of graphene surface systems containing surface defects, and achieved comparable accuracy with six orders of magnitude faster learning than when the GNN was trained from scratch.
期刊介绍:
Applied Physics Express (APEX) is a letters journal devoted solely to rapid dissemination of up-to-date and concise reports on new findings in applied physics. The motto of APEX is high scientific quality and prompt publication. APEX is a sister journal of the Japanese Journal of Applied Physics (JJAP) and is published by IOP Publishing Ltd on behalf of the Japan Society of Applied Physics (JSAP).