Materials property prediction for limited datasets enabled by feature selection and joint learning with MODNet

Abstract In order to make accurate predictions of material properties, current machine-learning approaches generally require large amounts of data, which are often not available in practice. In this work, MODNet, an all-round framework, is presented which relies on a feedforward neural network, the...

Descripción completa

Guardado en:

Detalles Bibliográficos
Autores principales:	Pierre-Paul De Breuck, Geoffroy Hautier, Gian-Marco Rignanese
Formato:	article
Lenguaje:	EN
Publicado:	Nature Portfolio 2021
Materias:	Materials of engineering and construction. Mechanics of materials TA401-492 Computer software QA76.75-76.765
Acceso en línea:	https://doaj.org/article/766aa52f573542b988fe2fc011e128b7
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

id	oai:doaj.org-article:766aa52f573542b988fe2fc011e128b7
record_format	dspace
spelling	oai:doaj.org-article:766aa52f573542b988fe2fc011e128b72021-12-02T15:57:19ZMaterials property prediction for limited datasets enabled by feature selection and joint learning with MODNet10.1038/s41524-021-00552-22057-3960https://doaj.org/article/766aa52f573542b988fe2fc011e128b72021-06-01T00:00:00Zhttps://doi.org/10.1038/s41524-021-00552-2https://doaj.org/toc/2057-3960Abstract In order to make accurate predictions of material properties, current machine-learning approaches generally require large amounts of data, which are often not available in practice. In this work, MODNet, an all-round framework, is presented which relies on a feedforward neural network, the selection of physically meaningful features, and when applicable, joint-learning. Next to being faster in terms of training time, this approach is shown to outperform current graph-network models on small datasets. In particular, the vibrational entropy at 305 K of crystals is predicted with a mean absolute test error of 0.009 meV/K/atom (four times lower than previous studies). Furthermore, joint learning reduces the test error compared to single-target learning and enables the prediction of multiple properties at once, such as temperature functions. Finally, the selection algorithm highlights the most important features and thus helps to understand the underlying physics.Pierre-Paul De BreuckGeoffroy HautierGian-Marco RignaneseNature PortfolioarticleMaterials of engineering and construction. Mechanics of materialsTA401-492Computer softwareQA76.75-76.765ENnpj Computational Materials, Vol 7, Iss 1, Pp 1-8 (2021)
institution	DOAJ
collection	DOAJ
language	EN
topic	Materials of engineering and construction. Mechanics of materials TA401-492 Computer software QA76.75-76.765
spellingShingle	Materials of engineering and construction. Mechanics of materials TA401-492 Computer software QA76.75-76.765 Pierre-Paul De Breuck Geoffroy Hautier Gian-Marco Rignanese Materials property prediction for limited datasets enabled by feature selection and joint learning with MODNet
description	Abstract In order to make accurate predictions of material properties, current machine-learning approaches generally require large amounts of data, which are often not available in practice. In this work, MODNet, an all-round framework, is presented which relies on a feedforward neural network, the selection of physically meaningful features, and when applicable, joint-learning. Next to being faster in terms of training time, this approach is shown to outperform current graph-network models on small datasets. In particular, the vibrational entropy at 305 K of crystals is predicted with a mean absolute test error of 0.009 meV/K/atom (four times lower than previous studies). Furthermore, joint learning reduces the test error compared to single-target learning and enables the prediction of multiple properties at once, such as temperature functions. Finally, the selection algorithm highlights the most important features and thus helps to understand the underlying physics.
format	article
author	Pierre-Paul De Breuck Geoffroy Hautier Gian-Marco Rignanese
author_facet	Pierre-Paul De Breuck Geoffroy Hautier Gian-Marco Rignanese
author_sort	Pierre-Paul De Breuck
title	Materials property prediction for limited datasets enabled by feature selection and joint learning with MODNet
title_short	Materials property prediction for limited datasets enabled by feature selection and joint learning with MODNet
title_full	Materials property prediction for limited datasets enabled by feature selection and joint learning with MODNet
title_fullStr	Materials property prediction for limited datasets enabled by feature selection and joint learning with MODNet
title_full_unstemmed	Materials property prediction for limited datasets enabled by feature selection and joint learning with MODNet
title_sort	materials property prediction for limited datasets enabled by feature selection and joint learning with modnet
publisher	Nature Portfolio
publishDate	2021
url	https://doaj.org/article/766aa52f573542b988fe2fc011e128b7
work_keys_str_mv	AT pierrepauldebreuck materialspropertypredictionforlimiteddatasetsenabledbyfeatureselectionandjointlearningwithmodnet AT geoffroyhautier materialspropertypredictionforlimiteddatasetsenabledbyfeatureselectionandjointlearningwithmodnet AT gianmarcorignanese materialspropertypredictionforlimiteddatasetsenabledbyfeatureselectionandjointlearningwithmodnet
_version_	1718385336655020032

Materials property prediction for limited datasets enabled by feature selection and joint learning with MODNet

Ejemplares similares