Semi-Supervised Classification on Non-Sparse Graphs Using Low-Rank Graph Convolutional Networks

TL;DR

This paper introduces low-rank graph convolutional networks and a reduced-order architecture to improve semi-supervised learning on large non-sparse graphs, achieving faster training and better accuracy.

Contribution

It proposes low-rank filters and a reduced-order GCN architecture, extending applicability to hypergraphs and enhancing efficiency and performance.

Findings

Significant runtime acceleration for non-sparse graphs

Improved accuracy over traditional GCNs on large datasets

Efficient hypergraph convolution implementation

Abstract

Graph Convolutional Networks (GCNs) have proven to be successful tools for semi-supervised learning on graph-based datasets. For sparse graphs, linear and polynomial filter functions have yielded impressive results. For large non-sparse graphs, however, network training and evaluation becomes prohibitively expensive. By introducing low-rank filters, we gain significant runtime acceleration and simultaneously improved accuracy. We further propose an architecture change mimicking techniques from Model Order Reduction in what we call a reduced-order GCN. Moreover, we present how our method can also be applied to hypergraph datasets and how hypergraph convolution can be implemented efficiently.