BARReL : Bottleneck Attention for Adversarial Robustness in Vision-Based Reinforcement Learning

Bykovets, Eugene; Metz, Yannick; El-Assady, Mennatallah; Keim, Daniel A.; Buhmann, Joachim M.

BARReL : Bottleneck Attention for Adversarial Robustness in Vision-Based Reinforcement Learning

Dateien

Zu diesem Dokument gibt es keine Dateien.

Datum

2022

Autor:innen

Bykovets, Eugene

Metz, Yannick

El-Assady, Mennatallah

Keim, Daniel A.

Buhmann, Joachim M.

ArXiv-ID

http://arxiv.org/abs/2208.10481

Sammlungen

Informatik und Informationswissenschaft

Publikationstyp

Preprint

Publikationsstatus

Published

Zusammenfassung

Robustness to adversarial perturbations has been explored in many areas of computer vision. This robustness is particularly relevant in vision-based reinforcement learning, as the actions of autonomous agents might be safety-critic or impactful in the real world. We investigate the susceptibility of vision-based reinforcement learning agents to gradient-based adversarial attacks and evaluate a potential defense. We observe that Bottleneck Attention Modules (BAM) included in CNN architectures can act as potential tools to increase robustness against adversarial attacks. We show how learned attention maps can be used to recover activations of a convolutional layer by restricting the spatial activations to salient regions. Across a number of RL environments, BAM-enhanced architectures show increased robustness during inference. Finally, we discuss potential future research directions.

Fachgebiet (DDC)

004 Informatik

Zitieren

ISO 690

BYKOVETS, Eugene, Yannick METZ, Mennatallah EL-ASSADY, Daniel A. KEIM, Joachim M. BUHMANN, 2022. BARReL : Bottleneck Attention for Adversarial Robustness in Vision-Based Reinforcement Learning

BibTex

@unpublished{Bykovets2022BARRe-58409,
  year={2022},
  title={BARReL : Bottleneck Attention for Adversarial Robustness in Vision-Based Reinforcement Learning},
  author={Bykovets, Eugene and Metz, Yannick and El-Assady, Mennatallah and Keim, Daniel A. and Buhmann, Joachim M.}
}

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/123456789/58409">
    <dc:contributor>Bykovets, Eugene</dc:contributor>
    <dc:creator>Keim, Daniel A.</dc:creator>
    <dc:date rdf:datatype="http://www.w3.org/2001/XMLSchema#dateTime">2022-08-29T14:25:15Z</dc:date>
    <dc:contributor>Buhmann, Joachim M.</dc:contributor>
    <dcterms:rights rdf:resource="https://rightsstatements.org/page/InC/1.0/"/>
    <dcterms:isPartOf rdf:resource="https://kops.uni-konstanz.de/server/rdf/resource/123456789/36"/>
    <bibo:uri rdf:resource="https://kops.uni-konstanz.de/handle/123456789/58409"/>
    <dc:creator>El-Assady, Mennatallah</dc:creator>
    <dc:contributor>Metz, Yannick</dc:contributor>
    <dc:rights>terms-of-use</dc:rights>
    <dc:creator>Buhmann, Joachim M.</dc:creator>
    <foaf:homepage rdf:resource="http://localhost:8080/"/>
    <dc:contributor>Keim, Daniel A.</dc:contributor>
    <dcterms:issued>2022</dcterms:issued>
    <dc:contributor>El-Assady, Mennatallah</dc:contributor>
    <dspace:isPartOfCollection rdf:resource="https://kops.uni-konstanz.de/server/rdf/resource/123456789/36"/>
    <void:sparqlEndpoint rdf:resource="http://localhost/fuseki/dspace/sparql"/>
    <dcterms:abstract xml:lang="eng">Robustness to adversarial perturbations has been explored in many areas of computer vision. This robustness is particularly relevant in vision-based reinforcement learning, as the actions of autonomous agents might be safety-critic or impactful in the real world. We investigate the susceptibility of vision-based reinforcement learning agents to gradient-based adversarial attacks and evaluate a potential defense. We observe that Bottleneck Attention Modules (BAM) included in CNN architectures can act as potential tools to increase robustness against adversarial attacks. We show how learned attention maps can be used to recover activations of a convolutional layer by restricting the spatial activations to salient regions. Across a number of RL environments, BAM-enhanced architectures show increased robustness during inference. Finally, we discuss potential future research directions.</dcterms:abstract>
    <dc:creator>Metz, Yannick</dc:creator>
    <dc:creator>Bykovets, Eugene</dc:creator>
    <dcterms:available rdf:datatype="http://www.w3.org/2001/XMLSchema#dateTime">2022-08-29T14:25:15Z</dcterms:available>
    <dcterms:title>BARReL : Bottleneck Attention for Adversarial Robustness in Vision-Based Reinforcement Learning</dcterms:title>
    <dc:language>eng</dc:language>
  </rdf:Description>
</rdf:RDF>

Universitätsbibliographie

Ja