No Cover Image

Conference contribution 256 views 75 downloads

Graph Convolutional Neural Network / Mike Edwards; Xianghua Xie

British Machine Vision Conference

Swansea University Author: Xie, Xianghua

Abstract

The benefit of localized features within the regular domain has given rise to the use of Convolutional Neural Networks (CNNs) in machine learning, with great proficiency in the image classification. The use of CNNs becomes problematic within the irregular spatial domain due to design and convolution...

Full description

Published in: British Machine Vision Conference
Published: 2016
URI: https://cronfa.swan.ac.uk/Record/cronfa32103
Tags: Add Tag
No Tags, Be the first to tag this record!
first_indexed 2017-02-25T05:04:37Z
last_indexed 2018-02-09T05:19:40Z
id cronfa32103
recordtype SURis
fullrecord <?xml version="1.0"?><rfc1807><datestamp>2017-03-23T12:27:29Z</datestamp><bib-version>v2</bib-version><id>32103</id><entry>2017-02-24</entry><title>Graph Convolutional Neural Network</title><alternativeTitle></alternativeTitle><author>Xianghua Xie</author><firstname>Xianghua</firstname><surname>Xie</surname><active>true</active><ORCID>0000-0002-2701-8660</ORCID><ethesisStudent>false</ethesisStudent><sid>b334d40963c7a2f435f06d2c26c74e11</sid><email>53b7e8cec1e3c035df428f36f80bdea5</email><emailaddr>ulOdsUw0nzyNlMFzZoDyVp320YwKTXZRCaAvm14NMEw=</emailaddr><date>2017-02-24</date><deptcode>SCS</deptcode><abstract>The benefit of localized features within the regular domain has given rise to the use of Convolutional Neural Networks (CNNs) in machine learning, with great proficiency in the image classification. The use of CNNs becomes problematic within the irregular spatial domain due to design and convolution of a kernel filter being non-trivial. One so- lution to this problem is to utilize graph signal processing techniques and the convolution theorem to perform convolutions on the graph of the irregular domain to obtain feature map responses to learnt filters. We propose graph convolution and pooling operators analogous to those in the regular domain. We also provide gradient calculations on the input data and spectral filters, which allow for the deep learning of an irregular spatial do- main problem. Signal filters take the form of spectral multipliers, applying convolution in the graph spectral domain. Applying smooth multipliers results in localized convo- lutions in the spatial domain, with smoother multipliers providing sharper feature maps. Algebraic Multigrid is presented as a graph pooling method, reducing the resolution of the graph through agglomeration of nodes between layers of the network. Evaluation of performance on the MNIST digit classification problem in both the regular and irregu- lar domain is presented, with comparison drawn to standard CNN. The proposed graph CNN provides a deep learning method for the irregular domains present in the machine learning community, obtaining 94.23% on the regular grid, and 94.96% on a spatially irregular subsampled MNIST.</abstract><type>Conference contribution</type><journal>British Machine Vision Conference</journal><volume/><journalNumber/><paginationStart/><paginationEnd/><publisher></publisher><placeOfPublication/><isbnPrint/><isbnElectronic/><issnPrint/><issnElectronic/><keywords>Convolutional Neural Network, Deep Learning, Machine Learning, Graph CNN</keywords><publishedDay>0</publishedDay><publishedMonth>9</publishedMonth><publishedYear>2016</publishedYear><publishedDate>2016-09-01</publishedDate><doi></doi><url></url><notes></notes><college>College of Science</college><department>Computer Science</department><CollegeCode>CSCI</CollegeCode><DepartmentCode>SCS</DepartmentCode><institution/><researchGroup>Visual Computing</researchGroup><supervisor/><sponsorsfunders/><grantnumber/><degreelevel/><degreename>None</degreename><lastEdited>2017-03-23T12:27:29Z</lastEdited><Created>2017-02-24T23:30:27Z</Created><path><level id="1">College of Science</level><level id="2">Computer Science</level></path><authors><author><firstname>Mike</firstname><surname>Edwards</surname><orcid/><order>1</order></author><author><firstname>Xianghua</firstname><surname>Xie</surname><orcid/><order>2</order></author></authors><documents><document><filename>0032103-24022017233147.pdf</filename><originalFilename>bmvc2016.pdf</originalFilename><uploaded>2017-02-24T23:31:47Z</uploaded><type>Output</type><contentLength>2266155</contentLength><contentType>application/pdf</contentType><version>AM</version><cronfaStatus>true</cronfaStatus><action>Published to Cronfa</action><actionDate>23/03/2017</actionDate><embargoDate>2017-02-24T00:00:00</embargoDate><documentNotes/><copyrightCorrect>true</copyrightCorrect><language>eng</language></document></documents></rfc1807>
spelling 2017-03-23T12:27:29Z v2 32103 2017-02-24 Graph Convolutional Neural Network Xianghua Xie Xianghua Xie true 0000-0002-2701-8660 false b334d40963c7a2f435f06d2c26c74e11 53b7e8cec1e3c035df428f36f80bdea5 ulOdsUw0nzyNlMFzZoDyVp320YwKTXZRCaAvm14NMEw= 2017-02-24 SCS The benefit of localized features within the regular domain has given rise to the use of Convolutional Neural Networks (CNNs) in machine learning, with great proficiency in the image classification. The use of CNNs becomes problematic within the irregular spatial domain due to design and convolution of a kernel filter being non-trivial. One so- lution to this problem is to utilize graph signal processing techniques and the convolution theorem to perform convolutions on the graph of the irregular domain to obtain feature map responses to learnt filters. We propose graph convolution and pooling operators analogous to those in the regular domain. We also provide gradient calculations on the input data and spectral filters, which allow for the deep learning of an irregular spatial do- main problem. Signal filters take the form of spectral multipliers, applying convolution in the graph spectral domain. Applying smooth multipliers results in localized convo- lutions in the spatial domain, with smoother multipliers providing sharper feature maps. Algebraic Multigrid is presented as a graph pooling method, reducing the resolution of the graph through agglomeration of nodes between layers of the network. Evaluation of performance on the MNIST digit classification problem in both the regular and irregu- lar domain is presented, with comparison drawn to standard CNN. The proposed graph CNN provides a deep learning method for the irregular domains present in the machine learning community, obtaining 94.23% on the regular grid, and 94.96% on a spatially irregular subsampled MNIST. Conference contribution British Machine Vision Conference Convolutional Neural Network, Deep Learning, Machine Learning, Graph CNN 0 9 2016 2016-09-01 College of Science Computer Science CSCI SCS Visual Computing None 2017-03-23T12:27:29Z 2017-02-24T23:30:27Z College of Science Computer Science Mike Edwards 1 Xianghua Xie 2 0032103-24022017233147.pdf bmvc2016.pdf 2017-02-24T23:31:47Z Output 2266155 application/pdf AM true Published to Cronfa 23/03/2017 2017-02-24T00:00:00 true eng
title Graph Convolutional Neural Network
spellingShingle Graph Convolutional Neural Network
Xie, Xianghua
title_short Graph Convolutional Neural Network
title_full Graph Convolutional Neural Network
title_fullStr Graph Convolutional Neural Network
title_full_unstemmed Graph Convolutional Neural Network
title_sort Graph Convolutional Neural Network
author_id_str_mv b334d40963c7a2f435f06d2c26c74e11
author_id_fullname_str_mv b334d40963c7a2f435f06d2c26c74e11_***_Xie, Xianghua
author Xie, Xianghua
author2 Mike Edwards
Xianghua Xie
format Conference contribution
container_title British Machine Vision Conference
publishDate 2016
institution Swansea University
college_str College of Science
hierarchytype
hierarchy_top_id collegeofscience
hierarchy_top_title College of Science
hierarchy_parent_id collegeofscience
hierarchy_parent_title College of Science
department_str Computer Science{{{_:::_}}}College of Science{{{_:::_}}}Computer Science
document_store_str 1
active_str 1
researchgroup_str Visual Computing
description The benefit of localized features within the regular domain has given rise to the use of Convolutional Neural Networks (CNNs) in machine learning, with great proficiency in the image classification. The use of CNNs becomes problematic within the irregular spatial domain due to design and convolution of a kernel filter being non-trivial. One so- lution to this problem is to utilize graph signal processing techniques and the convolution theorem to perform convolutions on the graph of the irregular domain to obtain feature map responses to learnt filters. We propose graph convolution and pooling operators analogous to those in the regular domain. We also provide gradient calculations on the input data and spectral filters, which allow for the deep learning of an irregular spatial do- main problem. Signal filters take the form of spectral multipliers, applying convolution in the graph spectral domain. Applying smooth multipliers results in localized convo- lutions in the spatial domain, with smoother multipliers providing sharper feature maps. Algebraic Multigrid is presented as a graph pooling method, reducing the resolution of the graph through agglomeration of nodes between layers of the network. Evaluation of performance on the MNIST digit classification problem in both the regular and irregu- lar domain is presented, with comparison drawn to standard CNN. The proposed graph CNN provides a deep learning method for the irregular domains present in the machine learning community, obtaining 94.23% on the regular grid, and 94.96% on a spatially irregular subsampled MNIST.
published_date 2016-09-01T13:38:49Z
_version_ 1625904675186278400
score 10.77982