Publikation:

A Robust Optimisation Perspective on Counterexample-Guided Repair of Neural Networks

Lade...
Vorschaubild

Dateien

Boetius_2-y8w0lfpv6ens3.pdf
Boetius_2-y8w0lfpv6ens3.pdfGröße: 759.61 KBDownloads: 10

Datum

2023

Herausgeber:innen

Kontakt

ISSN der Zeitschrift

Electronic ISSN

ISBN

Bibliografische Daten

Verlag

Schriftenreihe

Auflagebezeichnung

DOI (zitierfähiger Link)
ArXiv-ID

Internationale Patentnummer

Link zur Lizenz
oops

Angaben zur Forschungsförderung

Projekt

Open Access-Veröffentlichung
Open Access Green
Core Facility der Universität Konstanz

Gesperrt bis

Titel in einer weiteren Sprache

Publikationstyp
Beitrag zu einem Konferenzband
Publikationsstatus
Published

Erschienen in

Proceedings of the 40th International Conference on Machine Learning : ICML 2023. OpenReview, 2023

Zusammenfassung

Counterexample-guided repair aims at creating neural networks with mathematical safety guarantees, facilitating the application of neural networks in safety-critical domains. However, whether counterexample-guided repair is guaranteed to terminate remains an open question. We approach this question by showing that counterexample-guided repair can be viewed as a robust optimisation algorithm. While termination guarantees for neural network repair itself remain beyond our reach, we prove termination for more restrained machine learning models and disprove termination in a general setting. We empirically study the practical implications of our theoretical results, demonstrating the suitability of common verifiers and falsifiers for repair despite a disadvantageous theoretical result. Additionally, we use our theoretical insights to devise a novel algorithm for repairing linear regression models based on quadratic programming, surpassing existing approaches.

Zusammenfassung in einer weiteren Sprache

Fachgebiet (DDC)
004 Informatik

Schlagwörter

Konferenz

40th International Conference on Machine Learning : ICML 2023, 23. Juli 2023 - 29. Juli 2023, Honolulu, Hawaii
Rezension
undefined / . - undefined, undefined

Forschungsvorhaben

Organisationseinheiten

Zeitschriftenheft

Zugehörige Datensätze in KOPS

Zitieren

ISO 690BOETIUS, David, Stefan LEUE, Tobias SUTTER, 2023. A Robust Optimisation Perspective on Counterexample-Guided Repair of Neural Networks. 40th International Conference on Machine Learning : ICML 2023. Honolulu, Hawaii, 23. Juli 2023 - 29. Juli 2023. In: Proceedings of the 40th International Conference on Machine Learning : ICML 2023. OpenReview, 2023
BibTex
@inproceedings{Boetius2023Robus-70510,
  year={2023},
  title={A Robust Optimisation Perspective on Counterexample-Guided Repair of Neural Networks},
  url={https://openreview.net/forum?id=z3hnQh5UJd},
  publisher={OpenReview},
  booktitle={Proceedings of the 40th International Conference on Machine Learning : ICML 2023},
  author={Boetius, David and Leue, Stefan and Sutter, Tobias}
}
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/123456789/70510">
    <void:sparqlEndpoint rdf:resource="http://localhost/fuseki/dspace/sparql"/>
    <dc:creator>Sutter, Tobias</dc:creator>
    <dspace:hasBitstream rdf:resource="https://kops.uni-konstanz.de/bitstream/123456789/70510/1/Boetius_2-y8w0lfpv6ens3.pdf"/>
    <dcterms:title>A Robust Optimisation Perspective on Counterexample-Guided Repair of Neural Networks</dcterms:title>
    <dcterms:isPartOf rdf:resource="https://kops.uni-konstanz.de/server/rdf/resource/123456789/36"/>
    <bibo:uri rdf:resource="https://kops.uni-konstanz.de/handle/123456789/70510"/>
    <dc:creator>Boetius, David</dc:creator>
    <dspace:isPartOfCollection rdf:resource="https://kops.uni-konstanz.de/server/rdf/resource/123456789/36"/>
    <dc:contributor>Leue, Stefan</dc:contributor>
    <dc:language>eng</dc:language>
    <dcterms:issued>2023</dcterms:issued>
    <dcterms:hasPart rdf:resource="https://kops.uni-konstanz.de/bitstream/123456789/70510/1/Boetius_2-y8w0lfpv6ens3.pdf"/>
    <dc:contributor>Sutter, Tobias</dc:contributor>
    <dc:date rdf:datatype="http://www.w3.org/2001/XMLSchema#dateTime">2024-07-31T10:30:18Z</dc:date>
    <dc:creator>Leue, Stefan</dc:creator>
    <dcterms:abstract>Counterexample-guided repair aims at creating neural networks with mathematical safety guarantees, facilitating the application of neural networks in safety-critical domains. However, whether counterexample-guided repair is guaranteed to terminate remains an open question. We approach this question by showing that counterexample-guided repair can be viewed as a robust optimisation algorithm. While termination guarantees for neural network repair itself remain beyond our reach, we prove termination for more restrained machine learning models and disprove termination in a general setting. We empirically study the practical implications of our theoretical results, demonstrating the suitability of common verifiers and falsifiers for repair despite a disadvantageous theoretical result. Additionally, we use our theoretical insights to devise a novel algorithm for repairing linear regression models based on quadratic programming, surpassing existing approaches.</dcterms:abstract>
    <dcterms:available rdf:datatype="http://www.w3.org/2001/XMLSchema#dateTime">2024-07-31T10:30:18Z</dcterms:available>
    <dc:contributor>Boetius, David</dc:contributor>
    <foaf:homepage rdf:resource="http://localhost:8080/"/>
  </rdf:Description>
</rdf:RDF>

Interner Vermerk

xmlui.Submission.submit.DescribeStep.inputForms.label.kops_note_fromSubmitter

Kontakt

Prüfdatum der URL

2024-07-31

Prüfungsdatum der Dissertation

Finanzierungsart

Kommentar zur Publikation

Allianzlizenz
Corresponding Authors der Uni Konstanz vorhanden
Internationale Co-Autor:innen
Universitätsbibliographie
Ja
Begutachtet
Diese Publikation teilen