Geometric graphs from data to aid classification tasks with Graph Convolutional Networks
File(s)
Published version
Author(s)
Qian, Yifan
Expert, Paul
Panzarasa, Pietro
Barahona, Mauricio
Type
Journal Article
Abstract
Traditional classification tasks learn to assign samples to given classes based solely on sample features. This paradigm is evolving to include other sources of information, such as known relations between samples. Here, we show that, even if additional relational information is not available in the dataset, one can improve classification by constructing geometric graphs from the features themselves, and using them within a Graph Convolutional Network. The improvement in classification accuracy is maximized by graphs that capture sample similarity with relatively low edge density. We show that such feature-derived graphs increase the alignment of the data to the ground truth while improving class separation. We also demonstrate that the graphs can be made more efficient using spectral sparsification, which reduces the number of edges while still improving classification performance. We illustrate our findings using synthetic and real-world datasets from various scientific domains.
Date Issued
2021-04-09
Date Acceptance
2021-03-12
Citation
Patterns, 2021, 2 (4)
ISSN
2666-3899
Publisher
Cell Press
Journal / Book Title
Patterns
Volume
2
Issue
4
Copyright Statement
© 2021 The Authors. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/)
License URL
Sponsor
Engineering & Physical Science Research Council (EPSRC)
Grant Number
EP/N014529/1
Subjects
cs.LG
cs.LG
cs.SI
physics.soc-ph
stat.ML
Publication Status
Published
Article Number
ARTN 100237
Date Publish Online
2021-04-09