puc-header

Geometric Graphs from Data to Aid Classification Tasks with Graph Convolutional Networks

29 Pages Posted: 16 Oct 2020 Publication Status: Under Review

See all articles by Yifan Qian

Yifan Qian

Queen Mary University of London - School of Business and Management

Paul Expert

Imperial College London - School of Public Health

Pietro Panzarasa

Queen Mary University of London - School of Business and Management

Mauricio Barahona

Imperial College London - Department of Mathematics

More...

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 data set, 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 data sets from various scientific domains.

Keywords: classification tasks, machine learning, graph convolutional networks, graph neural networks, geometric deep learning, graph construction, graph sparsification, graph theory, network science, data science

Suggested Citation

Qian, Yifan and Expert, Paul and Panzarasa, Pietro and Barahona, Mauricio, Geometric Graphs from Data to Aid Classification Tasks with Graph Convolutional Networks. Available at SSRN: https://ssrn.com/abstract=3713501 or http://dx.doi.org/10.2139/ssrn.3713501
This is a paper under consideration at Cell Press and has not been peer-reviewed.

Yifan Qian

Queen Mary University of London - School of Business and Management ( email )

Mile End Rd
London, E1 4NS
United Kingdom

Paul Expert

Imperial College London - School of Public Health ( email )

London
United Kingdom

Pietro Panzarasa

Queen Mary University of London - School of Business and Management ( email )

Mile End Rd
London, E1 4NS
United Kingdom

Mauricio Barahona (Contact Author)

Imperial College London - Department of Mathematics ( email )

South Kensington Campus
Imperial College
LONDON, SW7 2AZ
United Kingdom

Click here to go to Cell.com

Paper statistics

Abstract Views
60
Downloads
6
PlumX Metrics