A Persistent WeisfeilerLehman Procedure for Graph Classifjcation - - PowerPoint PPT Presentation

a persistent weisfeiler lehman procedure for graph
SMART_READER_LITE
LIVE PREVIEW

A Persistent WeisfeilerLehman Procedure for Graph Classifjcation - - PowerPoint PPT Presentation

A Persistent WeisfeilerLehman Procedure for Graph Classifjcation Bastian Rieck Christian Bock Karsten Borgwardt Graph classifjcation Graph Potential labels A Persistent WeisfeilerLehman Procedure for Graph Classifjcation Bastian Rieck,


slide-1
SLIDE 1

A Persistent Weisfeiler–Lehman Procedure for Graph Classifjcation

Bastian Rieck Christian Bock Karsten Borgwardt

slide-2
SLIDE 2

Graph classifjcation

Graph Potential labels

A Persistent Weisfeiler–Lehman Procedure for Graph Classifjcation Bastian Rieck, Christian Bock, Karsten Borgwardt 11 June 2019

slide-3
SLIDE 3

Graph classifjcation

Neighbourhood aggregation

A Persistent Weisfeiler–Lehman Procedure for Graph Classifjcation Bastian Rieck, Christian Bock, Karsten Borgwardt 11 June 2019

slide-4
SLIDE 4

Our proposed method

“P-WL = WL + TDA”

A Persistent Weisfeiler–Lehman Procedure for Graph Classifjcation Bastian Rieck, Christian Bock, Karsten Borgwardt 11 June 2019

slide-5
SLIDE 5

Topological features in graphs

Connected components and cycles

Graph A connected component

1 2

Cycles

A Persistent Weisfeiler–Lehman Procedure for Graph Classifjcation Bastian Rieck, Christian Bock, Karsten Borgwardt 11 June 2019

slide-6
SLIDE 6

Finding topological features

Calculating the persistent homology of a weighted graph

Labelled graph

A Persistent Weisfeiler–Lehman Procedure for Graph Classifjcation Bastian Rieck, Christian Bock, Karsten Borgwardt 11 June 2019

slide-7
SLIDE 7

Finding topological features

Calculating the persistent homology of a weighted graph

Weisfeiler–Lehman aggregation

A Persistent Weisfeiler–Lehman Procedure for Graph Classifjcation Bastian Rieck, Christian Bock, Karsten Borgwardt 11 June 2019

slide-8
SLIDE 8

Finding topological features

Calculating the persistent homology of a weighted graph

Label multisets

A Persistent Weisfeiler–Lehman Procedure for Graph Classifjcation Bastian Rieck, Christian Bock, Karsten Borgwardt 11 June 2019

slide-9
SLIDE 9

Finding topological features

Calculating the persistent homology of a weighted graph

Weighted graph

A Persistent Weisfeiler–Lehman Procedure for Graph Classifjcation Bastian Rieck, Christian Bock, Karsten Borgwardt 11 June 2019

slide-10
SLIDE 10

Finding topological features

Calculating the persistent homology of a weighted graph

Weighted graph Graph fjltration Topological feature descriptor

A Persistent Weisfeiler–Lehman Procedure for Graph Classifjcation Bastian Rieck, Christian Bock, Karsten Borgwardt 11 June 2019

slide-11
SLIDE 11

Finding topological features

Calculating the persistent homology of a weighted graph

Weighted graph Graph fjltration Topological feature descriptor

A Persistent Weisfeiler–Lehman Procedure for Graph Classifjcation Bastian Rieck, Christian Bock, Karsten Borgwardt 11 June 2019

slide-12
SLIDE 12

Finding topological features

Calculating the persistent homology of a weighted graph

Weighted graph Graph fjltration Topological feature descriptor

A Persistent Weisfeiler–Lehman Procedure for Graph Classifjcation Bastian Rieck, Christian Bock, Karsten Borgwardt 11 June 2019

slide-13
SLIDE 13

Finding topological features

Calculating the persistent homology of a weighted graph

Weighted graph Graph fjltration Topological feature descriptor

A Persistent Weisfeiler–Lehman Procedure for Graph Classifjcation Bastian Rieck, Christian Bock, Karsten Borgwardt 11 June 2019

slide-14
SLIDE 14

Finding topological features

Calculating the persistent homology of a weighted graph

Weighted graph Graph fjltration Topological feature descriptor

A Persistent Weisfeiler–Lehman Procedure for Graph Classifjcation Bastian Rieck, Christian Bock, Karsten Borgwardt 11 June 2019

slide-15
SLIDE 15

Finding topological features

Calculating the persistent homology of a weighted graph

Weighted graph Graph fjltration Topological feature descriptor

A Persistent Weisfeiler–Lehman Procedure for Graph Classifjcation Bastian Rieck, Christian Bock, Karsten Borgwardt 11 June 2019

slide-16
SLIDE 16

Conclusion

P-WL… …is almost as effjcient as the Weisfeiler–Lehman kernel …exhibits favourable performance …outperforms “deep” graph kernels …shows the potential of topological data analysis (TDA)

Poster #224

A Persistent Weisfeiler–Lehman Procedure for Graph Classifjcation Bastian Rieck, Christian Bock, Karsten Borgwardt 11 June 2019