Layer-wise relevance propagation for deep neural network architectures

Alexander Binder; Sebastian Bach; Gregoire Montavon; Klaus Muller; Wojciech Samek

doi:10.1007/978-981-10-0557-2_87

Layer-wise relevance propagation for deep neural network architectures

Alexander Binder, Sebastian Bach, Gregoire Montavon, Klaus Muller, Wojciech Samek

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

112 Citations (Scopus)

Abstract

We present the application of layer-wise relevance propagation to several deep neural networks such as the BVLC reference neural net and googlenet trained on ImageNet and MIT Places datasets. Layer-wise relevance propagation is a method to compute scores for image pixels and image regions denoting the impact of the particular image region on the prediction of the classifier for one particular test image. We demonstrate the impact of different parameter settings on the resulting explanation.

Original language	English
Title of host publication	Lecture Notes in Electrical Engineering
Publisher	Springer Verlag
Pages	913-922
Number of pages	10
Volume	376
ISBN (Print)	9789811005565
DOIs	https://doi.org/10.1007/978-981-10-0557-2_87
Publication status	Published - 2016
Externally published	Yes
Event	International Conference on Information Science and Applications, ICISA 2016 - Minh City, Viet Nam Duration: 2016 Feb 15 → 2016 Feb 18

Publication series

Name	Lecture Notes in Electrical Engineering
Volume	376
ISSN (Print)	18761100
ISSN (Electronic)	18761119

Other

Other	International Conference on Information Science and Applications, ICISA 2016
Country/Territory	Viet Nam
City	Minh City
Period	16/2/15 → 16/2/18

Keywords

Deep neural networks
Non-linear explanations

ASJC Scopus subject areas

Industrial and Manufacturing Engineering

Access to Document

10.1007/978-981-10-0557-2_87

Cite this

Binder, A, Bach, S, Montavon, G, Muller, K & Samek, W 2016, Layer-wise relevance propagation for deep neural network architectures. in Lecture Notes in Electrical Engineering. vol. 376, Lecture Notes in Electrical Engineering, vol. 376, Springer Verlag, pp. 913-922, International Conference on Information Science and Applications, ICISA 2016, Minh City, Viet Nam, 16/2/15. https://doi.org/10.1007/978-981-10-0557-2_87

@inproceedings{8e24bfc6b6004092b5a42425545a9d92,

title = "Layer-wise relevance propagation for deep neural network architectures",

abstract = "We present the application of layer-wise relevance propagation to several deep neural networks such as the BVLC reference neural net and googlenet trained on ImageNet and MIT Places datasets. Layer-wise relevance propagation is a method to compute scores for image pixels and image regions denoting the impact of the particular image region on the prediction of the classifier for one particular test image. We demonstrate the impact of different parameter settings on the resulting explanation.",

keywords = "Deep neural networks, Non-linear explanations",

author = "Alexander Binder and Sebastian Bach and Gregoire Montavon and Klaus Muller and Wojciech Samek",

year = "2016",

doi = "10.1007/978-981-10-0557-2_87",

language = "English",

isbn = "9789811005565",

volume = "376",

series = "Lecture Notes in Electrical Engineering",

publisher = "Springer Verlag",

pages = "913--922",

booktitle = "Lecture Notes in Electrical Engineering",

note = "International Conference on Information Science and Applications, ICISA 2016 ; Conference date: 15-02-2016 Through 18-02-2016",

}

TY - GEN

T1 - Layer-wise relevance propagation for deep neural network architectures

AU - Binder, Alexander

AU - Bach, Sebastian

AU - Montavon, Gregoire

AU - Muller, Klaus

AU - Samek, Wojciech

PY - 2016

Y1 - 2016

N2 - We present the application of layer-wise relevance propagation to several deep neural networks such as the BVLC reference neural net and googlenet trained on ImageNet and MIT Places datasets. Layer-wise relevance propagation is a method to compute scores for image pixels and image regions denoting the impact of the particular image region on the prediction of the classifier for one particular test image. We demonstrate the impact of different parameter settings on the resulting explanation.

AB - We present the application of layer-wise relevance propagation to several deep neural networks such as the BVLC reference neural net and googlenet trained on ImageNet and MIT Places datasets. Layer-wise relevance propagation is a method to compute scores for image pixels and image regions denoting the impact of the particular image region on the prediction of the classifier for one particular test image. We demonstrate the impact of different parameter settings on the resulting explanation.

KW - Deep neural networks

KW - Non-linear explanations

UR - http://www.scopus.com/inward/record.url?scp=84959099772&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84959099772&partnerID=8YFLogxK

U2 - 10.1007/978-981-10-0557-2_87

DO - 10.1007/978-981-10-0557-2_87

M3 - Conference contribution

AN - SCOPUS:84959099772

SN - 9789811005565

VL - 376

T3 - Lecture Notes in Electrical Engineering

SP - 913

EP - 922

BT - Lecture Notes in Electrical Engineering

PB - Springer Verlag

T2 - International Conference on Information Science and Applications, ICISA 2016

Y2 - 15 February 2016 through 18 February 2016

ER -

Layer-wise relevance propagation for deep neural network architectures

Abstract

Publication series

Other

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this