Sight-seeing in the eyes of deep neural networks

Seyran Khademi; Xiangwei Shi; Tino Mager; Ronald Siebes; Carola Hein; Victor De Boer; Jan Van Gemert

doi:10.1109/eScience.2018.00125

Sight-seeing in the eyes of deep neural networks

Seyran Khademi, Xiangwei Shi, Tino Mager, Ronald Siebes, Carola Hein, Victor De Boer, Jan Van Gemert

Research output: Chapter in Book/Conference proceedings/Edited volume › Conference contribution › Scientific › peer-review

2 Citations (Scopus)

50 Downloads (Pure)

Abstract

We address the interpretability of convolutional neural networks (CNNs) for predicting a geo-location from an image. In a pilot experiment we classify images of Pittsburgh vs Tokyo and visualize the learned CNN filters. We found that varying the CNN architecture leads to variating in the visualized filters. This calls for further investigation of the effective parameters on the interpretability of CNNs.

Original language	English
Title of host publication	Proceedings - IEEE 14th International Conference on eScience, e-Science 2018
Editors	W. Hazeleger
Publisher	IEEE
Pages	407-408
Number of pages	2
ISBN (Electronic)	978-153869156-4
DOIs	https://doi.org/10.1109/eScience.2018.00125
Publication status	Published - 2018
Event	14th IEEE International Conference on eScience, e-Science 2018 - Amsterdam, Netherlands Duration: 29 Oct 2018 → 1 Nov 2018

Conference

Conference	14th IEEE International Conference on eScience, e-Science 2018
Country/Territory	Netherlands
City	Amsterdam
Period	29/10/18 → 1/11/18

Bibliographical note

Green Open Access added to TU Delft Institutional Repository ‘You share, we take care!’ – Taverne project https://www.openaccess.nl/en/you-share-we-take-care

Otherwise as indicated in the copyright section: the publisher is the copyright holder of this work and the author uses the Dutch legislation to make this work public.

Keywords

Classification
Convolutional neural network (CNN)
Interpretability
Place recognition
Visualization

Access to Document

10.1109/eScience.2018.00125

08588744Final published version, 748 KB

Cite this

@inproceedings{c7b3733a98484e55996f978b49c98a60,

title = "Sight-seeing in the eyes of deep neural networks",

abstract = "We address the interpretability of convolutional neural networks (CNNs) for predicting a geo-location from an image. In a pilot experiment we classify images of Pittsburgh vs Tokyo and visualize the learned CNN filters. We found that varying the CNN architecture leads to variating in the visualized filters. This calls for further investigation of the effective parameters on the interpretability of CNNs.",

keywords = "Classification, Convolutional neural network (CNN), Interpretability, Place recognition, Visualization",

author = "Seyran Khademi and Xiangwei Shi and Tino Mager and Ronald Siebes and Carola Hein and {De Boer}, Victor and {Van Gemert}, Jan",

note = "Green Open Access added to TU Delft Institutional Repository {\textquoteleft}You share, we take care!{\textquoteright} – Taverne project https://www.openaccess.nl/en/you-share-we-take-care Otherwise as indicated in the copyright section: the publisher is the copyright holder of this work and the author uses the Dutch legislation to make this work public.; 14th IEEE International Conference on eScience, e-Science 2018 ; Conference date: 29-10-2018 Through 01-11-2018",

year = "2018",

doi = "10.1109/eScience.2018.00125",

language = "English",

pages = "407--408",

editor = "W. Hazeleger",

booktitle = "Proceedings - IEEE 14th International Conference on eScience, e-Science 2018",

publisher = "IEEE",

address = "United States",

}

Khademi, S, Shi, X, Mager, T, Siebes, R, Hein, C, De Boer, V & Van Gemert, J 2018, Sight-seeing in the eyes of deep neural networks. in W Hazeleger (ed.), Proceedings - IEEE 14th International Conference on eScience, e-Science 2018., 8588744, IEEE, pp. 407-408, 14th IEEE International Conference on eScience, e-Science 2018, Amsterdam, Netherlands, 29/10/18. https://doi.org/10.1109/eScience.2018.00125

TY - GEN

T1 - Sight-seeing in the eyes of deep neural networks

AU - Khademi, Seyran

AU - Shi, Xiangwei

AU - Mager, Tino

AU - Siebes, Ronald

AU - Hein, Carola

AU - De Boer, Victor

AU - Van Gemert, Jan

N1 - Green Open Access added to TU Delft Institutional Repository ‘You share, we take care!’ – Taverne project https://www.openaccess.nl/en/you-share-we-take-care Otherwise as indicated in the copyright section: the publisher is the copyright holder of this work and the author uses the Dutch legislation to make this work public.

PY - 2018

Y1 - 2018

N2 - We address the interpretability of convolutional neural networks (CNNs) for predicting a geo-location from an image. In a pilot experiment we classify images of Pittsburgh vs Tokyo and visualize the learned CNN filters. We found that varying the CNN architecture leads to variating in the visualized filters. This calls for further investigation of the effective parameters on the interpretability of CNNs.

AB - We address the interpretability of convolutional neural networks (CNNs) for predicting a geo-location from an image. In a pilot experiment we classify images of Pittsburgh vs Tokyo and visualize the learned CNN filters. We found that varying the CNN architecture leads to variating in the visualized filters. This calls for further investigation of the effective parameters on the interpretability of CNNs.

KW - Classification

KW - Convolutional neural network (CNN)

KW - Interpretability

KW - Place recognition

KW - Visualization

UR - http://www.scopus.com/inward/record.url?scp=85061364522&partnerID=8YFLogxK

U2 - 10.1109/eScience.2018.00125

DO - 10.1109/eScience.2018.00125

M3 - Conference contribution

SP - 407

EP - 408

BT - Proceedings - IEEE 14th International Conference on eScience, e-Science 2018

A2 - Hazeleger, W.

PB - IEEE

T2 - 14th IEEE International Conference on eScience, e-Science 2018

Y2 - 29 October 2018 through 1 November 2018

ER -

Sight-seeing in the eyes of deep neural networks

Abstract

Conference

Bibliographical note

Keywords

Access to Document

Other files and links

Fingerprint

Cite this