ub.xmlui.mirage2.page-structure.muninLogoub.xmlui.mirage2.page-structure.openResearchArchiveLogo
    • EnglishEnglish
    • norsknorsk
  • Velg spraakEnglish 
    • EnglishEnglish
    • norsknorsk
  • Administration/UB
View Item 
  •   Home
  • Universitetsbiblioteket
  • Artikler, rapporter og annet (UB)
  • View Item
  •   Home
  • Universitetsbiblioteket
  • Artikler, rapporter og annet (UB)
  • View Item
JavaScript is disabled for your browser. Some features of this site may not work without it.

Transferability of graph neural networks: An extended graphon approach

Permanent link
https://hdl.handle.net/10037/33224
DOI
https://doi.org/10.1016/j.acha.2022.11.008
Thumbnail
View/Open
article.pdf (456.4Kb)
Accepted manuscript version (PDF)
Date
2022-11-28
Type
Journal article
Tidsskriftartikkel
Peer reviewed

Author
Maskey, Sohir; Levie, Ron; Kutyniok, Gitta Astrid Hildegard
Abstract
We study spectral graph convolutional neural networks (GCNNs), where filters are defined as continuous functions of the graph shift operator (GSO) through functional calculus. A spectral GCNN is not tailored to one specific graph and can be transferred between different graphs. It is hence important to study the GCNN transferability: the capacity of the network to have approximately the same repercussion on different graphs that represent the same phenomenon. Transferability ensures that GCNNs trained on certain graphs generalize if the graphs in the test set represent the same phenomena as the graphs in the training set.

In this paper, we consider a model of transferability based on graphon analysis. Graphons are limit objects of graphs, and, in the graph paradigm, two graphs represent the same phenomenon if both approximate the same graphon. Our main contributions can be summarized as follows: 1) we prove that any fixed GCNN with continuous filters is transferable under graphs that approximate the same graphon, 2) we prove transferability for graphs that approximate unbounded graphon shift operators, which are defined in this paper, and 3) we obtain non-asymptotic approximation results, proving linear stability of GCNNs. This extends current state-of-the-art results which show asymptotic transferability for polynomial filters under graphs that approximate bounded graphons.

Publisher
Elsevier
Citation
Maskey, Levie, Kutyniok. Transferability of graph neural networks: An extended graphon approach. Applied and Computational Harmonic Analysis. 2023;63:48-83
Metadata
Show full item record
Collections
  • Artikler, rapporter og annet (UB) [3245]
Copyright 2023 The Author(s)

Browse

Browse all of MuninCommunities & CollectionsAuthor listTitlesBy Issue DateBrowse this CollectionAuthor listTitlesBy Issue Date
Login

Statistics

View Usage Statistics
UiT

Munin is powered by DSpace

UiT The Arctic University of Norway
The University Library
uit.no/ub - munin@ub.uit.no

Accessibility statement (Norwegian only)