Comparing transferability in neural network ...
Type de document :
Article dans une revue scientifique
URL permanente :
Titre :
Comparing transferability in neural network approaches and linear models for machine-learning interaction potentials
Auteur(s) :
Kandy, Akshay Krishna Ammothum [Auteur]
Centre d'élaboration de matériaux et d'études structurales [CEMES]
Rossi, Kevin [Auteur]
Ecole Polytechnique Fédérale de Lausanne [EPFL]
Raulin-Foissac, Alexis [Auteur]
Centre d'élaboration de matériaux et d'études structurales [CEMES]
Laurens, Gaétan [Auteur]
Institut Lumière Matière [Villeurbanne] [ILM]
Lam, Julien [Auteur]
Unité Matériaux et Transformations (UMET) - UMR 8207
Centre d'élaboration de matériaux et d'études structurales [CEMES]
Rossi, Kevin [Auteur]
Ecole Polytechnique Fédérale de Lausanne [EPFL]
Raulin-Foissac, Alexis [Auteur]
Centre d'élaboration de matériaux et d'études structurales [CEMES]
Laurens, Gaétan [Auteur]
Institut Lumière Matière [Villeurbanne] [ILM]
Lam, Julien [Auteur]
Unité Matériaux et Transformations (UMET) - UMR 8207
Titre de la revue :
Physical Review B
Nom court de la revue :
Phys. Rev. B
Numéro :
107
Pagination :
174106
Éditeur :
American Physical Society (APS)
Date de publication :
2023-05-10
ISSN :
2469-9950
Discipline(s) HAL :
Physique [physics]/Matière Condensée [cond-mat]/Science des matériaux [cond-mat.mtrl-sci]
Planète et Univers [physics]/Sciences de la Terre
Planète et Univers [physics]/Sciences de la Terre
Résumé en anglais : [en]
Atomic simulations using machine learning interatomic potential (MLIP) have gained a lot of popularity owing to their accuracy in comparison to conventional empirical potentials. However, the transferability of MLIP to ...
Lire la suite >Atomic simulations using machine learning interatomic potential (MLIP) have gained a lot of popularity owing to their accuracy in comparison to conventional empirical potentials. However, the transferability of MLIP to systems outside the training set poses a significant challenge. Here, we compare the transferability of three MLIP approaches: (i) neural network potentials (NNP), (ii) physical LassoLars interactions potential (PLIP) and (iii) linear potentials with Belher-Parrinello descriptors, trained over a small but diverse configuration of zinc oxide polymorphs. We compared the obtained models with density functional theory reference results for physical properties including bulk lattice parameters, surface energies, and vibrational density of states and showed the superiority of both NNP and PLIP models. However, the NNP model performed poorly when compared to the other two linear models for the structural optimization of nanoparticles and molecular dynamics simulation of liquid phases, which are systems outside the training set. While providing less accurate prediction for solid Zinc Oxides phases, both linear models appear more transferable than NNP when testing for nanoscale systems and liquid phases. Our results are finally rationalized by a combination of different statistical analysis including spread in force evaluation, information imbalance, convex hull calculation, and density in descriptor space.Lire moins >
Lire la suite >Atomic simulations using machine learning interatomic potential (MLIP) have gained a lot of popularity owing to their accuracy in comparison to conventional empirical potentials. However, the transferability of MLIP to systems outside the training set poses a significant challenge. Here, we compare the transferability of three MLIP approaches: (i) neural network potentials (NNP), (ii) physical LassoLars interactions potential (PLIP) and (iii) linear potentials with Belher-Parrinello descriptors, trained over a small but diverse configuration of zinc oxide polymorphs. We compared the obtained models with density functional theory reference results for physical properties including bulk lattice parameters, surface energies, and vibrational density of states and showed the superiority of both NNP and PLIP models. However, the NNP model performed poorly when compared to the other two linear models for the structural optimization of nanoparticles and molecular dynamics simulation of liquid phases, which are systems outside the training set. While providing less accurate prediction for solid Zinc Oxides phases, both linear models appear more transferable than NNP when testing for nanoscale systems and liquid phases. Our results are finally rationalized by a combination of different statistical analysis including spread in force evaluation, information imbalance, convex hull calculation, and density in descriptor space.Lire moins >
Langue :
Anglais
Audience :
Internationale
Vulgarisation :
Non
Établissement(s) :
Université de Lille
CNRS
INRAE
ENSCL
CNRS
INRAE
ENSCL
Collections :
Équipe(s) de recherche :
Plasticité
Date de dépôt :
2023-12-21T07:48:04Z
2023-12-22T10:42:52Z
2023-12-22T10:42:52Z
Fichiers
- 2303_NNvsPLIP.pdf
- Version soumise (preprint)
- Accès libre
- Accéder au document