Controlling explanatory heatmap resolution and semantics via decomposition depth

Sebastian Bach; Alexander Binder; Klaus Muller; Wojciech Samek

doi:10.1109/ICIP.2016.7532763

Controlling explanatory heatmap resolution and semantics via decomposition depth

Sebastian Bach, Alexander Binder, Klaus Muller, Wojciech Samek

Department of Brain and Cognitive Engineering

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

15 Citations (Scopus)

Abstract

We present an application of the Layer-wise Relevance Propagation (LRP) algorithm to state of the art deep convolutional neural networks and Fisher Vector classifiers to compare the image perception and prediction strategies of both classifiers with the use of visualized heatmaps. Layer-wise Relevance Propagation (LRP) is a method to compute scores for individual components of an input image, denoting their contribution to the prediction of the classifier for one particular test point. We demonstrate the impact of different choices of decomposition cut-off points during the LRP-process, controlling the resolution and semantics of the heatmap on test images from the PASCAL VOC 2007 test data set.

Original language	English
Title of host publication	2016 IEEE International Conference on Image Processing, ICIP 2016 - Proceedings
Publisher	IEEE Computer Society
Pages	2271-2275
Number of pages	5
Volume	2016-August
ISBN (Electronic)	9781467399616
DOIs	https://doi.org/10.1109/ICIP.2016.7532763
Publication status	Published - 2016 Aug 3
Event	23rd IEEE International Conference on Image Processing, ICIP 2016 - Phoenix, United States Duration: 2016 Sept 25 → 2016 Sept 28

Other

Other	23rd IEEE International Conference on Image Processing, ICIP 2016
Country/Territory	United States
City	Phoenix
Period	16/9/25 → 16/9/28

Keywords

Explaining Classifiers
Heatmapping

ASJC Scopus subject areas

Software
Computer Vision and Pattern Recognition
Signal Processing

Access to Document

10.1109/ICIP.2016.7532763

Cite this

Controlling explanatory heatmap resolution and semantics via decomposition depth. / Bach, Sebastian; Binder, Alexander; Muller, Klaus et al.
2016 IEEE International Conference on Image Processing, ICIP 2016 - Proceedings. Vol. 2016-August IEEE Computer Society, 2016. p. 2271-2275 7532763.

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

Bach, S, Binder, A, Muller, K & Samek, W 2016, Controlling explanatory heatmap resolution and semantics via decomposition depth. in 2016 IEEE International Conference on Image Processing, ICIP 2016 - Proceedings. vol. 2016-August, 7532763, IEEE Computer Society, pp. 2271-2275, 23rd IEEE International Conference on Image Processing, ICIP 2016, Phoenix, United States, 16/9/25. https://doi.org/10.1109/ICIP.2016.7532763

@inproceedings{376a8854e8df41f28b8238d8f398d7cd,

title = "Controlling explanatory heatmap resolution and semantics via decomposition depth",

abstract = "We present an application of the Layer-wise Relevance Propagation (LRP) algorithm to state of the art deep convolutional neural networks and Fisher Vector classifiers to compare the image perception and prediction strategies of both classifiers with the use of visualized heatmaps. Layer-wise Relevance Propagation (LRP) is a method to compute scores for individual components of an input image, denoting their contribution to the prediction of the classifier for one particular test point. We demonstrate the impact of different choices of decomposition cut-off points during the LRP-process, controlling the resolution and semantics of the heatmap on test images from the PASCAL VOC 2007 test data set.",

keywords = "Explaining Classifiers, Heatmapping",

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

year = "2016",

month = aug,

day = "3",

doi = "10.1109/ICIP.2016.7532763",

language = "English",

volume = "2016-August",

pages = "2271--2275",

booktitle = "2016 IEEE International Conference on Image Processing, ICIP 2016 - Proceedings",

publisher = "IEEE Computer Society",

note = "23rd IEEE International Conference on Image Processing, ICIP 2016 ; Conference date: 25-09-2016 Through 28-09-2016",

}

TY - GEN

T1 - Controlling explanatory heatmap resolution and semantics via decomposition depth

AU - Bach, Sebastian

AU - Binder, Alexander

AU - Muller, Klaus

AU - Samek, Wojciech

PY - 2016/8/3

Y1 - 2016/8/3

N2 - We present an application of the Layer-wise Relevance Propagation (LRP) algorithm to state of the art deep convolutional neural networks and Fisher Vector classifiers to compare the image perception and prediction strategies of both classifiers with the use of visualized heatmaps. Layer-wise Relevance Propagation (LRP) is a method to compute scores for individual components of an input image, denoting their contribution to the prediction of the classifier for one particular test point. We demonstrate the impact of different choices of decomposition cut-off points during the LRP-process, controlling the resolution and semantics of the heatmap on test images from the PASCAL VOC 2007 test data set.

AB - We present an application of the Layer-wise Relevance Propagation (LRP) algorithm to state of the art deep convolutional neural networks and Fisher Vector classifiers to compare the image perception and prediction strategies of both classifiers with the use of visualized heatmaps. Layer-wise Relevance Propagation (LRP) is a method to compute scores for individual components of an input image, denoting their contribution to the prediction of the classifier for one particular test point. We demonstrate the impact of different choices of decomposition cut-off points during the LRP-process, controlling the resolution and semantics of the heatmap on test images from the PASCAL VOC 2007 test data set.

KW - Explaining Classifiers

KW - Heatmapping

UR - http://www.scopus.com/inward/record.url?scp=85006759512&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85006759512&partnerID=8YFLogxK

U2 - 10.1109/ICIP.2016.7532763

DO - 10.1109/ICIP.2016.7532763

M3 - Conference contribution

AN - SCOPUS:85006759512

VL - 2016-August

SP - 2271

EP - 2275

BT - 2016 IEEE International Conference on Image Processing, ICIP 2016 - Proceedings

PB - IEEE Computer Society

T2 - 23rd IEEE International Conference on Image Processing, ICIP 2016

Y2 - 25 September 2016 through 28 September 2016

ER -

Controlling explanatory heatmap resolution and semantics via decomposition depth

Abstract

Other

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this