GLAM: Graph Learning by Modeling Affinity to Labeled Nodes for Graph Neural Networks

Vijay Lingam, Arun Iyer, Rahul Ragesh

Keywords: Graph Neural Networks, Graph Learning

TL;DR: We propose a simple graph learning approach by modeling label-affinity that is up to 70x faster to train and gives performance lifts up to 1.5%.

Abstract: We propose GLAM, a semi-supervised graph learning method for cases when there are no graphs available. This approach learns a graph as a convex combination of the unsupervised k-Nearest Neighbor graph and a supervised label-affinity graph. The latter graph directly captures all the nodes' label-affinity with the labeled nodes, i.e., how likely a node has the same label as the labeled nodes. Our experiments show that GLAM gives close to or better performance (up to $\sim$1.5\%), while being simpler and faster (up to 70x) to train, than state-of-the-art graph learning methods. We also demonstrate the importance of individual components and contrast them with the state-of-the-art methods.