Identifying domains of applicability of machine learning models for materials science

Sutton, Christopher; Boley, Mario; Ghiringhelli, Luca M.; Rupp, Matthias; Vreeken, Jilles; Scheffler, Matthias

Identifying domains of applicability of machine learning models for materials science

Dateien

Sutton_2-1auzvjbeeo0k13.pdfGröße: 5.29 MBDownloads: 315

Datum

2020

Autor:innen

Sutton, Christopher

Boley, Mario

Ghiringhelli, Luca M.

Rupp, Matthias

Vreeken, Jilles

Scheffler, Matthias

Open Access-Veröffentlichung

Open Access Gold

Sammlungen

Informatik und Informationswissenschaft

Publikationstyp

Zeitschriftenartikel

Publikationsstatus

Published

Erschienen in

Nature communications. Nature Publishing Group. 2020, 11(1), 4428. eISSN 2041-1723. Available under: doi: 10.1038/s41467-020-17112-9

Zusammenfassung

Although machine learning (ML) models promise to substantially accelerate the discovery of novel materials, their performance is often still insufficient to draw reliable conclusions. Improved ML models are therefore actively researched, but their design is currently guided mainly by monitoring the average model test error. This can render different models indistinguishable although their performance differs substantially across materials, or it can make a model appear generally insufficient while it actually works well in specific sub-domains. Here, we present a method, based on subgroup discovery, for detecting domains of applicability (DA) of models within a materials class. The utility of this approach is demonstrated by analyzing three state-of-the-art ML models for predicting the formation energy of transparent conducting oxides. We find that, despite having a mutually indistinguishable and unsatisfactory average error, the models have DAs with distinctive features and notably improved performance.

Fachgebiet (DDC)

004 Informatik

Zitieren

ISO 690

SUTTON, Christopher, Mario BOLEY, Luca M. GHIRINGHELLI, Matthias RUPP, Jilles VREEKEN, Matthias SCHEFFLER, 2020. Identifying domains of applicability of machine learning models for materials science. In: Nature communications. Nature Publishing Group. 2020, 11(1), 4428. eISSN 2041-1723. Available under: doi: 10.1038/s41467-020-17112-9

BibTex

@article{Sutton2020Ident-51258,
  year={2020},
  doi={10.1038/s41467-020-17112-9},
  title={Identifying domains of applicability of machine learning models for materials science},
  number={1},
  volume={11},
  journal={Nature communications},
  author={Sutton, Christopher and Boley, Mario and Ghiringhelli, Luca M. and Rupp, Matthias and Vreeken, Jilles and Scheffler, Matthias},
  note={Article Number: 4428}
}

RDF

<rdf:RDF
    xmlns:dcterms="http://purl.org/dc/terms/"
    xmlns:dc="http://purl.org/dc/elements/1.1/"
    xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
    xmlns:bibo="http://purl.org/ontology/bibo/"
    xmlns:dspace="http://digital-repositories.org/ontologies/dspace/0.1.0#"
    xmlns:foaf="http://xmlns.com/foaf/0.1/"
    xmlns:void="http://rdfs.org/ns/void#"
    xmlns:xsd="http://www.w3.org/2001/XMLSchema#" > 
  <rdf:Description rdf:about="https://kops.uni-konstanz.de/server/rdf/resource/123456789/51258">
    <dc:contributor>Sutton, Christopher</dc:contributor>
    <dcterms:rights rdf:resource="http://creativecommons.org/licenses/by/4.0/"/>
    <dspace:hasBitstream rdf:resource="https://kops.uni-konstanz.de/bitstream/123456789/51258/1/Sutton_2-1auzvjbeeo0k13.pdf"/>
    <dc:creator>Ghiringhelli, Luca M.</dc:creator>
    <dc:creator>Vreeken, Jilles</dc:creator>
    <dc:contributor>Rupp, Matthias</dc:contributor>
    <dc:language>eng</dc:language>
    <dcterms:available rdf:datatype="http://www.w3.org/2001/XMLSchema#dateTime">2020-10-07T14:05:45Z</dcterms:available>
    <dc:date rdf:datatype="http://www.w3.org/2001/XMLSchema#dateTime">2020-10-07T14:05:45Z</dc:date>
    <foaf:homepage rdf:resource="http://localhost:8080/"/>
    <void:sparqlEndpoint rdf:resource="http://localhost/fuseki/dspace/sparql"/>
    <dc:rights>Attribution 4.0 International</dc:rights>
    <dc:contributor>Ghiringhelli, Luca M.</dc:contributor>
    <dcterms:isPartOf rdf:resource="https://kops.uni-konstanz.de/server/rdf/resource/123456789/36"/>
    <dspace:isPartOfCollection rdf:resource="https://kops.uni-konstanz.de/server/rdf/resource/123456789/36"/>
    <dc:creator>Rupp, Matthias</dc:creator>
    <dc:contributor>Boley, Mario</dc:contributor>
    <dc:contributor>Scheffler, Matthias</dc:contributor>
    <dcterms:abstract xml:lang="eng">Although machine learning (ML) models promise to substantially accelerate the discovery of novel materials, their performance is often still insufficient to draw reliable conclusions. Improved ML models are therefore actively researched, but their design is currently guided mainly by monitoring the average model test error. This can render different models indistinguishable although their performance differs substantially across materials, or it can make a model appear generally insufficient while it actually works well in specific sub-domains. Here, we present a method, based on subgroup discovery, for detecting domains of applicability (DA) of models within a materials class. The utility of this approach is demonstrated by analyzing three state-of-the-art ML models for predicting the formation energy of transparent conducting oxides. We find that, despite having a mutually indistinguishable and unsatisfactory average error, the models have DAs with distinctive features and notably improved performance.</dcterms:abstract>
    <dc:creator>Scheffler, Matthias</dc:creator>
    <dcterms:issued>2020</dcterms:issued>
    <bibo:uri rdf:resource="https://kops.uni-konstanz.de/handle/123456789/51258"/>
    <dc:contributor>Vreeken, Jilles</dc:contributor>
    <dc:creator>Sutton, Christopher</dc:creator>
    <dc:creator>Boley, Mario</dc:creator>
    <dcterms:hasPart rdf:resource="https://kops.uni-konstanz.de/bitstream/123456789/51258/1/Sutton_2-1auzvjbeeo0k13.pdf"/>
    <dcterms:title>Identifying domains of applicability of machine learning models for materials science</dcterms:title>
  </rdf:Description>
</rdf:RDF>

Universitätsbibliographie

Ja

Begutachtet

Ja