Scaling law of Sim2Real transfer learning in expanding computational materials databases for real-world predictions

Abstract To address the challenge of limited experimental materials data, extensive physical property databases are being developed based on high-throughput computational experiments, such as molecular dynamics simulations. Previous studies have shown that fine-tuning a predictor pretrained on a com...

Full description

Saved in:
Bibliographic Details
Main Authors: Shunya Minami, Yoshihiro Hayashi, Stephen Wu, Kenji Fukumizu, Hiroki Sugisawa, Masashi Ishii, Isao Kuwajima, Kazuya Shiratori, Ryo Yoshida
Format: Article
Language:English
Published: Nature Portfolio 2025-05-01
Series:npj Computational Materials
Online Access:https://doi.org/10.1038/s41524-025-01606-5
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850207483736358912
author Shunya Minami
Yoshihiro Hayashi
Stephen Wu
Kenji Fukumizu
Hiroki Sugisawa
Masashi Ishii
Isao Kuwajima
Kazuya Shiratori
Ryo Yoshida
author_facet Shunya Minami
Yoshihiro Hayashi
Stephen Wu
Kenji Fukumizu
Hiroki Sugisawa
Masashi Ishii
Isao Kuwajima
Kazuya Shiratori
Ryo Yoshida
author_sort Shunya Minami
collection DOAJ
description Abstract To address the challenge of limited experimental materials data, extensive physical property databases are being developed based on high-throughput computational experiments, such as molecular dynamics simulations. Previous studies have shown that fine-tuning a predictor pretrained on a computational database to a real system can result in models with outstanding generalization capabilities compared to learning from scratch. This study demonstrates the scaling law of simulation-to-real (Sim2Real) transfer learning for several machine learning tasks in materials science. Case studies of three prediction tasks for polymers and inorganic materials reveal that the prediction error on real systems decreases according to a power-law as the size of the computational data increases. Observing the scaling behavior offers various insights for database development, such as determining the sample size necessary to achieve a desired performance, identifying equivalent sample sizes for physical and computational experiments, and guiding the design of data production protocols for downstream real-world tasks.
format Article
id doaj-art-feff0b14968a41c3b30b68553019f464
institution OA Journals
issn 2057-3960
language English
publishDate 2025-05-01
publisher Nature Portfolio
record_format Article
series npj Computational Materials
spelling doaj-art-feff0b14968a41c3b30b68553019f4642025-08-20T02:10:31ZengNature Portfolionpj Computational Materials2057-39602025-05-0111111010.1038/s41524-025-01606-5Scaling law of Sim2Real transfer learning in expanding computational materials databases for real-world predictionsShunya Minami0Yoshihiro Hayashi1Stephen Wu2Kenji Fukumizu3Hiroki Sugisawa4Masashi Ishii5Isao Kuwajima6Kazuya Shiratori7Ryo Yoshida8The Institute of Statistical Mathematics, Research Organization of Information and SystemsThe Institute of Statistical Mathematics, Research Organization of Information and SystemsThe Institute of Statistical Mathematics, Research Organization of Information and SystemsThe Institute of Statistical Mathematics, Research Organization of Information and SystemsScience & Innovation Center, Mitsubishi Chemical CorporationResearch and Service Division of Materials Data and Integrated System, National Institute for Materials ScienceResearch and Service Division of Materials Data and Integrated System, National Institute for Materials ScienceScience & Innovation Center, Mitsubishi Chemical CorporationThe Institute of Statistical Mathematics, Research Organization of Information and SystemsAbstract To address the challenge of limited experimental materials data, extensive physical property databases are being developed based on high-throughput computational experiments, such as molecular dynamics simulations. Previous studies have shown that fine-tuning a predictor pretrained on a computational database to a real system can result in models with outstanding generalization capabilities compared to learning from scratch. This study demonstrates the scaling law of simulation-to-real (Sim2Real) transfer learning for several machine learning tasks in materials science. Case studies of three prediction tasks for polymers and inorganic materials reveal that the prediction error on real systems decreases according to a power-law as the size of the computational data increases. Observing the scaling behavior offers various insights for database development, such as determining the sample size necessary to achieve a desired performance, identifying equivalent sample sizes for physical and computational experiments, and guiding the design of data production protocols for downstream real-world tasks.https://doi.org/10.1038/s41524-025-01606-5
spellingShingle Shunya Minami
Yoshihiro Hayashi
Stephen Wu
Kenji Fukumizu
Hiroki Sugisawa
Masashi Ishii
Isao Kuwajima
Kazuya Shiratori
Ryo Yoshida
Scaling law of Sim2Real transfer learning in expanding computational materials databases for real-world predictions
npj Computational Materials
title Scaling law of Sim2Real transfer learning in expanding computational materials databases for real-world predictions
title_full Scaling law of Sim2Real transfer learning in expanding computational materials databases for real-world predictions
title_fullStr Scaling law of Sim2Real transfer learning in expanding computational materials databases for real-world predictions
title_full_unstemmed Scaling law of Sim2Real transfer learning in expanding computational materials databases for real-world predictions
title_short Scaling law of Sim2Real transfer learning in expanding computational materials databases for real-world predictions
title_sort scaling law of sim2real transfer learning in expanding computational materials databases for real world predictions
url https://doi.org/10.1038/s41524-025-01606-5
work_keys_str_mv AT shunyaminami scalinglawofsim2realtransferlearninginexpandingcomputationalmaterialsdatabasesforrealworldpredictions
AT yoshihirohayashi scalinglawofsim2realtransferlearninginexpandingcomputationalmaterialsdatabasesforrealworldpredictions
AT stephenwu scalinglawofsim2realtransferlearninginexpandingcomputationalmaterialsdatabasesforrealworldpredictions
AT kenjifukumizu scalinglawofsim2realtransferlearninginexpandingcomputationalmaterialsdatabasesforrealworldpredictions
AT hirokisugisawa scalinglawofsim2realtransferlearninginexpandingcomputationalmaterialsdatabasesforrealworldpredictions
AT masashiishii scalinglawofsim2realtransferlearninginexpandingcomputationalmaterialsdatabasesforrealworldpredictions
AT isaokuwajima scalinglawofsim2realtransferlearninginexpandingcomputationalmaterialsdatabasesforrealworldpredictions
AT kazuyashiratori scalinglawofsim2realtransferlearninginexpandingcomputationalmaterialsdatabasesforrealworldpredictions
AT ryoyoshida scalinglawofsim2realtransferlearninginexpandingcomputationalmaterialsdatabasesforrealworldpredictions