Robust Generalization despite Distribution Shift via Minimum Discriminating Information

Sutter, Tobias; Krause, Andreas; Kuhn, Daniel

Robust Generalization despite Distribution Shift via Minimum Discriminating Information

Dateien

Zu diesem Dokument gibt es keine Dateien.

Datum

2021

Autor:innen

Sutter, Tobias

Krause, Andreas

Kuhn, Daniel

Link zur Lizenz

Urheberrechtlich geschützt

Sammlungen

Informatik und Informationswissenschaft

Publikationstyp

Beitrag zu einem Konferenzband

Publikationsstatus

Published

Erschienen in

Advances in Neural Information Processing Systems 34 pre-proceedings (NeurIPS 2021) / Ranzato, Marc'Aurelio; Beygelzimer, Alina; Dauphin, Yann et al. (Hrsg.). - San Diego, CA : Neural Information Processing Systems Foundation, 2021

Zusammenfassung

Training models that perform well under distribution shifts is a central challenge in machine learning. In this paper, we introduce a modeling framework where, in addition to training data, we have partial structural knowledge of the shifted test distribution. We employ the principle of minimum discriminating information to embed the available prior knowledge, and use distributionally robust optimization to account for uncertainty due to the limited samples. By leveraging large deviation results, we obtain explicit generalization bounds with respect to the unknown shifted distribution. Lastly, we demonstrate the versatility of our framework by demonstrating it on two rather distinct applications: (1) training classifiers on systematically biased data and (2) off-policy evaluation in Markov Decision Processes.

Fachgebiet (DDC)

004 Informatik

Konferenz

NeurIPS 2021 : 35th Conference on Neural Information Processing Systems (online), 6. Dez. 2021 - 14. Dez. 2021

Zitieren

ISO 690

SUTTER, Tobias, Andreas KRAUSE, Daniel KUHN, 2021. Robust Generalization despite Distribution Shift via Minimum Discriminating Information. NeurIPS 2021 : 35th Conference on Neural Information Processing Systems (online), 6. Dez. 2021 - 14. Dez. 2021. In: RANZATO, Marc'Aurelio, ed., Alina BEYGELZIMER, ed., Yann DAUPHIN, ed. and others. Advances in Neural Information Processing Systems 34 pre-proceedings (NeurIPS 2021). San Diego, CA:Neural Information Processing Systems Foundation

BibTex

@inproceedings{Sutter2021Robus-55736,
  year={2021},
  title={Robust Generalization despite Distribution Shift via Minimum Discriminating Information},
  url={https://proceedings.neurips.cc/paper/2021/hash/f86890095c957e9b949d11d15f0d0cd5-Abstract.html},
  publisher={Neural Information Processing Systems Foundation},
  address={San Diego, CA},
  booktitle={Advances in Neural Information Processing Systems 34 pre-proceedings (NeurIPS 2021)},
  editor={Ranzato, Marc'Aurelio and Beygelzimer, Alina and Dauphin, Yann},
  author={Sutter, Tobias and Krause, Andreas and Kuhn, Daniel}
}

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/123456789/55736">
    <dc:contributor>Kuhn, Daniel</dc:contributor>
    <dc:creator>Kuhn, Daniel</dc:creator>
    <bibo:uri rdf:resource="https://kops.uni-konstanz.de/handle/123456789/55736"/>
    <dc:rights>terms-of-use</dc:rights>
    <dc:creator>Krause, Andreas</dc:creator>
    <dcterms:available rdf:datatype="http://www.w3.org/2001/XMLSchema#dateTime">2021-12-02T12:27:08Z</dcterms:available>
    <dc:creator>Sutter, Tobias</dc:creator>
    <void:sparqlEndpoint rdf:resource="http://localhost/fuseki/dspace/sparql"/>
    <dcterms:isPartOf rdf:resource="https://kops.uni-konstanz.de/server/rdf/resource/123456789/36"/>
    <dc:language>eng</dc:language>
    <dc:contributor>Sutter, Tobias</dc:contributor>
    <dcterms:title>Robust Generalization despite Distribution Shift via Minimum Discriminating Information</dcterms:title>
    <foaf:homepage rdf:resource="http://localhost:8080/"/>
    <dcterms:issued>2021</dcterms:issued>
    <dcterms:abstract xml:lang="eng">Training models that perform well under distribution shifts is a central challenge in machine learning. In this paper, we introduce a modeling framework where, in addition to training data, we have partial structural knowledge of the shifted test distribution. We employ the principle of minimum discriminating information to embed the available prior knowledge, and use distributionally robust optimization to account for uncertainty due to the limited samples. By leveraging large deviation results, we obtain explicit generalization bounds with respect to the unknown shifted distribution. Lastly, we demonstrate the versatility of our framework by demonstrating it on two rather distinct applications: (1) training classifiers on systematically biased data and (2) off-policy evaluation in Markov Decision Processes.</dcterms:abstract>
    <dc:date rdf:datatype="http://www.w3.org/2001/XMLSchema#dateTime">2021-12-02T12:27:08Z</dc:date>
    <dspace:isPartOfCollection rdf:resource="https://kops.uni-konstanz.de/server/rdf/resource/123456789/36"/>
    <dcterms:rights rdf:resource="https://rightsstatements.org/page/InC/1.0/"/>
    <dc:contributor>Krause, Andreas</dc:contributor>
  </rdf:Description>
</rdf:RDF>

URL der Originalveröffentl.

https://proceedings.neurips.cc/paper/2021/hash/f86890095c957e9b949d11d15f0d0cd5-Abstract.html

Prüfdatum der URL

2021-11-23

Universitätsbibliographie

Ja