GraN: An efficient gradient-norm based detector for adversarial and misclassified examples

Julia Lust; Alexandru P. Condurache

GraN: An efficient gradient-norm based detector for adversarial and misclassified examples

Julia Lust, Alexandru P. Condurache

Institute of Signal Processing

Abstract

Deep neural networks (DNNs) are vulnerable to adversarial examples and other data perturbations. Especially in safety critical applications of DNNs, it is therefore crucial to detect misclassified samples. The current state-of-the-art detection methods require either significantly more runtime or more parameters than the original network itself. This paper therefore proposes GraN, a time- and parameter-efficient method that is easily adaptable to any DNN. GraN is based on the layer-wise norm of the DNN's gradient regarding the loss of the current input-output combination, which can be computed via backpropagation. GraN achieves state-of-the-art performance on numerous problem set-ups.

Original language	English
Number of pages	6
Publication status	Published - 10.2020
Event	European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning 2020 - Brügge, Belgium Duration: 02.10.2020 → 04.10.2020

Conference

Conference	European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning 2020
Abbreviated title	ESANN 2020
Country/Territory	Belgium
City	Brügge
Period	02.10.20 → 04.10.20

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

Access to Document

https://arxiv.org/pdf/2004.09179.pdf

Cite this

@conference{078743a0ca60445eb6bd3bd960c929db,

title = "GraN: An efficient gradient-norm based detector for adversarial and misclassified examples",

abstract = "Deep neural networks (DNNs) are vulnerable to adversarial examples and other data perturbations. Especially in safety critical applications of DNNs, it is therefore crucial to detect misclassified samples. The current state-of-the-art detection methods require either significantly more runtime or more parameters than the original network itself. This paper therefore proposes GraN, a time- and parameter-efficient method that is easily adaptable to any DNN. GraN is based on the layer-wise norm of the DNN's gradient regarding the loss of the current input-output combination, which can be computed via backpropagation. GraN achieves state-of-the-art performance on numerous problem set-ups.",

author = "Julia Lust and Condurache, {Alexandru P.}",

note = "Publisher Copyright: {\textcopyright} ESANN 2020 - Proceedings, 28th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning. Copyright: Copyright 2020 Elsevier B.V., All rights reserved.; European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning 2020 ; Conference date: 02-10-2020 Through 04-10-2020",

year = "2020",

month = oct,

language = "English",

}

TY - CONF

T1 - GraN: An efficient gradient-norm based detector for adversarial and misclassified examples

AU - Lust, Julia

AU - Condurache, Alexandru P.

N1 - Publisher Copyright: © ESANN 2020 - Proceedings, 28th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning. Copyright: Copyright 2020 Elsevier B.V., All rights reserved.

PY - 2020/10

Y1 - 2020/10

N2 - Deep neural networks (DNNs) are vulnerable to adversarial examples and other data perturbations. Especially in safety critical applications of DNNs, it is therefore crucial to detect misclassified samples. The current state-of-the-art detection methods require either significantly more runtime or more parameters than the original network itself. This paper therefore proposes GraN, a time- and parameter-efficient method that is easily adaptable to any DNN. GraN is based on the layer-wise norm of the DNN's gradient regarding the loss of the current input-output combination, which can be computed via backpropagation. GraN achieves state-of-the-art performance on numerous problem set-ups.

AB - Deep neural networks (DNNs) are vulnerable to adversarial examples and other data perturbations. Especially in safety critical applications of DNNs, it is therefore crucial to detect misclassified samples. The current state-of-the-art detection methods require either significantly more runtime or more parameters than the original network itself. This paper therefore proposes GraN, a time- and parameter-efficient method that is easily adaptable to any DNN. GraN is based on the layer-wise norm of the DNN's gradient regarding the loss of the current input-output combination, which can be computed via backpropagation. GraN achieves state-of-the-art performance on numerous problem set-ups.

UR - https://www.scopus.com/record/display.uri?eid=2-s2.0-85098374646&origin=inward&txGid=48a2ae9ee1d07d795a44a0abd664946d

M3 - Conference Papers

T2 - European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning 2020

Y2 - 2 October 2020 through 4 October 2020

ER -

GraN: An efficient gradient-norm based detector for adversarial and misclassified examples

Abstract

Conference

UN SDGs

Access to Document

Other files and links

Fingerprint

Cite this