ænet-PyTorch: A GPU-supported implementation for machine learning atomic potentials training

Jon López-Zorrilla; Xabier M. Aretxabaleta; In Won Yeu; Iñigo Etxebarria; Hegoi Manzano; Nongnuch Artrith

doi:10.1063/5.0146803

ænet-PyTorch: A GPU-supported implementation for machine learning atomic potentials training

Jon López-Zorrilla
, Xabier M. Aretxabaleta
, In Won Yeu
, Iñigo Etxebarria
, Hegoi Manzano
, Nongnuch Artrith^*

^*Corresponding author for this work

University of the Basque Country
Columbia University

Research output: Contribution to journal › Article › Academic › peer-review

Abstract

In this work, we present ænet-PyTorch, a PyTorch-based implementation for training artificial neural network-based machine learning interatomic potentials. Developed as an extension of the atomic energy network (ænet), ænet-PyTorch provides access to all the tools included in ænet for the application and usage of the potentials. The package has been designed as an alternative to the internal training capabilities of ænet, leveraging the power of graphic processing units to facilitate direct training on forces in addition to energies. This leads to a substantial reduction of the training time by one to two orders of magnitude compared to the central processing unit implementation, enabling direct training on forces for systems beyond small molecules. Here, we demonstrate the main features of ænet-PyTorch and show its performance on open databases. Our results show that training on all the force information within a dataset is not necessary, and including between 10% and 20% of the force information is sufficient to achieve optimally accurate interatomic potentials with the least computational resources.

Original language	English
Article number	164105
Pages (from-to)	1-10
Number of pages	10
Journal	Journal of Chemical Physics
Volume	158
Issue number	16
DOIs	https://doi.org/10.1063/5.0146803
Publication status	Published - 28 Apr 2023

Bibliographical note

Funding Information:
This work was supported by the “Departamento de Educación, Política Lingüística y Cultura del Gobierno Vasco” (IT1458-22), the “Ministerio de Ciencia e Innovación” (Grant No. PID2019-106644GB-I00), and the Project HPC-EUROPA3 (Grant No. INFRAIA-2016-1-730897), with the support of the EC Research Innovation Action under the H2020 Programme. The authors acknowledge technical and human support provided by SGIker (UPV/EHU/ERDF, EU) and the Duch National e-Infrastructure and the SURF Cooperative for computational resources (National Supercomputer Snellius). J.L.-Z. acknowledges financial support from the Basque Country Government (PRE_2019_1_0025). N.A. acknowledges funding from the Bayer AG Life Science Collaboration (“!AIQU”).

Funding Information:
This work was supported by the “Departamento de Educación, Política Lingüística y Cultura del Gobierno Vasco” (IT1458-22), the “Ministerio de Ciencia e Innovación” (Grant No. PID2019-106644GB-I00), and the Project HPC-EUROPA3 (Grant No. INFRAIA-2016-1-730897), with the support of the EC Research Innovation Action under the H2020 Programme. The authors acknowledge technical and human support provided by SGIker (UPV/EHU/ERDF, EU) and the Duch National e-Infrastructure and the SURF Cooperative for computational resources (National Supercomputer Snellius). J.L.-Z. acknowledges financial support from the Basque Country Government (PRE_2019_1_0025). N.A. acknowledges funding from the Bayer AG Life Science Collaboration (“!AIQU”).

Publisher Copyright:
© 2023 Author(s).

Funding

This work was supported by the “Departamento de Educación, Política Lingüística y Cultura del Gobierno Vasco” (IT1458-22), the “Ministerio de Ciencia e Innovación” (Grant No. PID2019-106644GB-I00), and the Project HPC-EUROPA3 (Grant No. INFRAIA-2016-1-730897), with the support of the EC Research Innovation Action under the H2020 Programme. The authors acknowledge technical and human support provided by SGIker (UPV/EHU/ERDF, EU) and the Duch National e-Infrastructure and the SURF Cooperative for computational resources (National Supercomputer Snellius). J.L.-Z. acknowledges financial support from the Basque Country Government (PRE_2019_1_0025). N.A. acknowledges funding from the Bayer AG Life Science Collaboration (“!AIQU”). This work was supported by the “Departamento de Educación, Política Lingüística y Cultura del Gobierno Vasco” (IT1458-22), the “Ministerio de Ciencia e Innovación” (Grant No. PID2019-106644GB-I00), and the Project HPC-EUROPA3 (Grant No. INFRAIA-2016-1-730897), with the support of the EC Research Innovation Action under the H2020 Programme. The authors acknowledge technical and human support provided by SGIker (UPV/EHU/ERDF, EU) and the Duch National e-Infrastructure and the SURF Cooperative for computational resources (National Supercomputer Snellius). J.L.-Z. acknowledges financial support from the Basque Country Government (PRE_2019_1_0025). N.A. acknowledges funding from the Bayer AG Life Science Collaboration (“!AIQU”).

Funders	Funder number
Departamento de Educacion, Politica Lingueistica y Cultura del Gobierno Vasco	IT1458-22
Ministerio de Ciencia e Innovacion	PID2019-106644GB-I00
Project HPC-EUROPA3	INFRAIA-2016-1-730897
EC Research Innovation Action under the H2020 Programme
Basque Country Government	PRE_2019_1_0025
Bayer AG Life Science Collaboration ("!AIQU")

Access to Document

10.1063/5.0146803Licence: CC BY

164105_1_5.0146803Final published version, 4.19 MBLicence: CC BY

Cite this

@article{b0cbacce955c4d8db291b4ac908d9975,

title = "{\ae}net-PyTorch: A GPU-supported implementation for machine learning atomic potentials training",

abstract = "In this work, we present {\ae}net-PyTorch, a PyTorch-based implementation for training artificial neural network-based machine learning interatomic potentials. Developed as an extension of the atomic energy network ({\ae}net), {\ae}net-PyTorch provides access to all the tools included in {\ae}net for the application and usage of the potentials. The package has been designed as an alternative to the internal training capabilities of {\ae}net, leveraging the power of graphic processing units to facilitate direct training on forces in addition to energies. This leads to a substantial reduction of the training time by one to two orders of magnitude compared to the central processing unit implementation, enabling direct training on forces for systems beyond small molecules. Here, we demonstrate the main features of {\ae}net-PyTorch and show its performance on open databases. Our results show that training on all the force information within a dataset is not necessary, and including between 10\% and 20\% of the force information is sufficient to achieve optimally accurate interatomic potentials with the least computational resources.",

author = "Jon L{\'o}pez-Zorrilla and Aretxabaleta, \{Xabier M.\} and Yeu, \{In Won\} and I{\~n}igo Etxebarria and Hegoi Manzano and Nongnuch Artrith",

note = "Funding Information: This work was supported by the {\textquotedblleft}Departamento de Educaci{\'o}n, Pol{\'i}tica Ling{\"u}{\'i}stica y Cultura del Gobierno Vasco{\textquotedblright} (IT1458-22), the {\textquotedblleft}Ministerio de Ciencia e Innovaci{\'o}n{\textquotedblright} (Grant No. PID2019-106644GB-I00), and the Project HPC-EUROPA3 (Grant No. INFRAIA-2016-1-730897), with the support of the EC Research Innovation Action under the H2020 Programme. The authors acknowledge technical and human support provided by SGIker (UPV/EHU/ERDF, EU) and the Duch National e-Infrastructure and the SURF Cooperative for computational resources (National Supercomputer Snellius). J.L.-Z. acknowledges financial support from the Basque Country Government (PRE\_2019\_1\_0025). N.A. acknowledges funding from the Bayer AG Life Science Collaboration ({\textquotedblleft}!AIQU{\textquotedblright}). Funding Information: This work was supported by the {\textquotedblleft}Departamento de Educaci{\'o}n, Pol{\'i}tica Ling{\"u}{\'i}stica y Cultura del Gobierno Vasco{\textquotedblright} (IT1458-22), the {\textquotedblleft}Ministerio de Ciencia e Innovaci{\'o}n{\textquotedblright} (Grant No. PID2019-106644GB-I00), and the Project HPC-EUROPA3 (Grant No. INFRAIA-2016-1-730897), with the support of the EC Research Innovation Action under the H2020 Programme. The authors acknowledge technical and human support provided by SGIker (UPV/EHU/ERDF, EU) and the Duch National e-Infrastructure and the SURF Cooperative for computational resources (National Supercomputer Snellius). J.L.-Z. acknowledges financial support from the Basque Country Government (PRE\_2019\_1\_0025). N.A. acknowledges funding from the Bayer AG Life Science Collaboration ({\textquotedblleft}!AIQU{\textquotedblright}). Publisher Copyright: {\textcopyright} 2023 Author(s).",

year = "2023",

month = apr,

day = "28",

doi = "10.1063/5.0146803",

language = "English",

volume = "158",

pages = "1--10",

journal = "Journal of Chemical Physics",

issn = "0021-9606",

publisher = "American Institute of Physics",

number = "16",

}

TY - JOUR

T1 - ænet-PyTorch

T2 - A GPU-supported implementation for machine learning atomic potentials training

AU - López-Zorrilla, Jon

AU - Aretxabaleta, Xabier M.

AU - Yeu, In Won

AU - Etxebarria, Iñigo

AU - Manzano, Hegoi

AU - Artrith, Nongnuch

N1 - Funding Information: This work was supported by the “Departamento de Educación, Política Lingüística y Cultura del Gobierno Vasco” (IT1458-22), the “Ministerio de Ciencia e Innovación” (Grant No. PID2019-106644GB-I00), and the Project HPC-EUROPA3 (Grant No. INFRAIA-2016-1-730897), with the support of the EC Research Innovation Action under the H2020 Programme. The authors acknowledge technical and human support provided by SGIker (UPV/EHU/ERDF, EU) and the Duch National e-Infrastructure and the SURF Cooperative for computational resources (National Supercomputer Snellius). J.L.-Z. acknowledges financial support from the Basque Country Government (PRE_2019_1_0025). N.A. acknowledges funding from the Bayer AG Life Science Collaboration (“!AIQU”). Funding Information: This work was supported by the “Departamento de Educación, Política Lingüística y Cultura del Gobierno Vasco” (IT1458-22), the “Ministerio de Ciencia e Innovación” (Grant No. PID2019-106644GB-I00), and the Project HPC-EUROPA3 (Grant No. INFRAIA-2016-1-730897), with the support of the EC Research Innovation Action under the H2020 Programme. The authors acknowledge technical and human support provided by SGIker (UPV/EHU/ERDF, EU) and the Duch National e-Infrastructure and the SURF Cooperative for computational resources (National Supercomputer Snellius). J.L.-Z. acknowledges financial support from the Basque Country Government (PRE_2019_1_0025). N.A. acknowledges funding from the Bayer AG Life Science Collaboration (“!AIQU”). Publisher Copyright: © 2023 Author(s).

PY - 2023/4/28

Y1 - 2023/4/28

N2 - In this work, we present ænet-PyTorch, a PyTorch-based implementation for training artificial neural network-based machine learning interatomic potentials. Developed as an extension of the atomic energy network (ænet), ænet-PyTorch provides access to all the tools included in ænet for the application and usage of the potentials. The package has been designed as an alternative to the internal training capabilities of ænet, leveraging the power of graphic processing units to facilitate direct training on forces in addition to energies. This leads to a substantial reduction of the training time by one to two orders of magnitude compared to the central processing unit implementation, enabling direct training on forces for systems beyond small molecules. Here, we demonstrate the main features of ænet-PyTorch and show its performance on open databases. Our results show that training on all the force information within a dataset is not necessary, and including between 10% and 20% of the force information is sufficient to achieve optimally accurate interatomic potentials with the least computational resources.

AB - In this work, we present ænet-PyTorch, a PyTorch-based implementation for training artificial neural network-based machine learning interatomic potentials. Developed as an extension of the atomic energy network (ænet), ænet-PyTorch provides access to all the tools included in ænet for the application and usage of the potentials. The package has been designed as an alternative to the internal training capabilities of ænet, leveraging the power of graphic processing units to facilitate direct training on forces in addition to energies. This leads to a substantial reduction of the training time by one to two orders of magnitude compared to the central processing unit implementation, enabling direct training on forces for systems beyond small molecules. Here, we demonstrate the main features of ænet-PyTorch and show its performance on open databases. Our results show that training on all the force information within a dataset is not necessary, and including between 10% and 20% of the force information is sufficient to achieve optimally accurate interatomic potentials with the least computational resources.

UR - https://www.scopus.com/pages/publications/85153687480

U2 - 10.1063/5.0146803

DO - 10.1063/5.0146803

M3 - Article

C2 - 37096855

AN - SCOPUS:85153687480

SN - 0021-9606

VL - 158

SP - 1

EP - 10

JO - Journal of Chemical Physics

JF - Journal of Chemical Physics

IS - 16

M1 - 164105

ER -

ænet-PyTorch: A GPU-supported implementation for machine learning atomic potentials training

Abstract

Bibliographical note

Funding

Access to Document

Other files and links

Fingerprint

Cite this