Enhancing materials property prediction by leveraging computational and experimental data using deep transfer learning

Dipendra Jha, Kamal Choudhary, Francesca Tavazza, Wei keng Liao, Alok Choudhary, Carelyn Campbell, Ankit Agrawal*

*Corresponding author for this work

Research output: Contribution to journalArticle

Abstract

The current predictive modeling techniques applied to Density Functional Theory (DFT) computations have helped accelerate the process of materials discovery by providing significantly faster methods to scan materials candidates, thereby reducing the search space for future DFT computations and experiments. However, in addition to prediction error against DFT-computed properties, such predictive models also inherit the DFT-computation discrepancies against experimentally measured properties. To address this challenge, we demonstrate that using deep transfer learning, existing large DFT-computational data sets (such as the Open Quantum Materials Database (OQMD)) can be leveraged together with other smaller DFT-computed data sets as well as available experimental observations to build robust prediction models. We build a highly accurate model for predicting formation energy of materials from their compositions; using an experimental data set of 1 , 963 observations, the proposed approach yields a mean absolute error (MAE) of 0.06 eV/atom, which is significantly better than existing machine learning (ML) prediction modeling based on DFT computations and is comparable to the MAE of DFT-computation itself.

Original languageEnglish (US)
Article number5316
JournalNature communications
Volume10
Issue number1
DOIs
StatePublished - Dec 1 2019

Fingerprint

learning
Density functional theory
Materials properties
density functional theory
predictions
Databases
machine learning
Datasets
Transfer (Psychology)
energy of formation
Learning systems
Atoms
Chemical analysis
atoms
Machine Learning
Experiments

ASJC Scopus subject areas

  • Chemistry(all)
  • Biochemistry, Genetics and Molecular Biology(all)
  • Physics and Astronomy(all)

Cite this

@article{4c773e57122e4f0993590522f1ac2bb5,
title = "Enhancing materials property prediction by leveraging computational and experimental data using deep transfer learning",
abstract = "The current predictive modeling techniques applied to Density Functional Theory (DFT) computations have helped accelerate the process of materials discovery by providing significantly faster methods to scan materials candidates, thereby reducing the search space for future DFT computations and experiments. However, in addition to prediction error against DFT-computed properties, such predictive models also inherit the DFT-computation discrepancies against experimentally measured properties. To address this challenge, we demonstrate that using deep transfer learning, existing large DFT-computational data sets (such as the Open Quantum Materials Database (OQMD)) can be leveraged together with other smaller DFT-computed data sets as well as available experimental observations to build robust prediction models. We build a highly accurate model for predicting formation energy of materials from their compositions; using an experimental data set of 1 , 963 observations, the proposed approach yields a mean absolute error (MAE) of 0.06 eV/atom, which is significantly better than existing machine learning (ML) prediction modeling based on DFT computations and is comparable to the MAE of DFT-computation itself.",
author = "Dipendra Jha and Kamal Choudhary and Francesca Tavazza and Liao, {Wei keng} and Alok Choudhary and Carelyn Campbell and Ankit Agrawal",
year = "2019",
month = "12",
day = "1",
doi = "10.1038/s41467-019-13297-w",
language = "English (US)",
volume = "10",
journal = "Nature Communications",
issn = "2041-1723",
publisher = "Nature Publishing Group",
number = "1",

}

Enhancing materials property prediction by leveraging computational and experimental data using deep transfer learning. / Jha, Dipendra; Choudhary, Kamal; Tavazza, Francesca; Liao, Wei keng; Choudhary, Alok; Campbell, Carelyn; Agrawal, Ankit.

In: Nature communications, Vol. 10, No. 1, 5316, 01.12.2019.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Enhancing materials property prediction by leveraging computational and experimental data using deep transfer learning

AU - Jha, Dipendra

AU - Choudhary, Kamal

AU - Tavazza, Francesca

AU - Liao, Wei keng

AU - Choudhary, Alok

AU - Campbell, Carelyn

AU - Agrawal, Ankit

PY - 2019/12/1

Y1 - 2019/12/1

N2 - The current predictive modeling techniques applied to Density Functional Theory (DFT) computations have helped accelerate the process of materials discovery by providing significantly faster methods to scan materials candidates, thereby reducing the search space for future DFT computations and experiments. However, in addition to prediction error against DFT-computed properties, such predictive models also inherit the DFT-computation discrepancies against experimentally measured properties. To address this challenge, we demonstrate that using deep transfer learning, existing large DFT-computational data sets (such as the Open Quantum Materials Database (OQMD)) can be leveraged together with other smaller DFT-computed data sets as well as available experimental observations to build robust prediction models. We build a highly accurate model for predicting formation energy of materials from their compositions; using an experimental data set of 1 , 963 observations, the proposed approach yields a mean absolute error (MAE) of 0.06 eV/atom, which is significantly better than existing machine learning (ML) prediction modeling based on DFT computations and is comparable to the MAE of DFT-computation itself.

AB - The current predictive modeling techniques applied to Density Functional Theory (DFT) computations have helped accelerate the process of materials discovery by providing significantly faster methods to scan materials candidates, thereby reducing the search space for future DFT computations and experiments. However, in addition to prediction error against DFT-computed properties, such predictive models also inherit the DFT-computation discrepancies against experimentally measured properties. To address this challenge, we demonstrate that using deep transfer learning, existing large DFT-computational data sets (such as the Open Quantum Materials Database (OQMD)) can be leveraged together with other smaller DFT-computed data sets as well as available experimental observations to build robust prediction models. We build a highly accurate model for predicting formation energy of materials from their compositions; using an experimental data set of 1 , 963 observations, the proposed approach yields a mean absolute error (MAE) of 0.06 eV/atom, which is significantly better than existing machine learning (ML) prediction modeling based on DFT computations and is comparable to the MAE of DFT-computation itself.

UR - http://www.scopus.com/inward/record.url?scp=85075525386&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85075525386&partnerID=8YFLogxK

U2 - 10.1038/s41467-019-13297-w

DO - 10.1038/s41467-019-13297-w

M3 - Article

C2 - 31757948

AN - SCOPUS:85075525386

VL - 10

JO - Nature Communications

JF - Nature Communications

SN - 2041-1723

IS - 1

M1 - 5316

ER -