Transferability for Graph Convolutional Networks

Christian Koke, Abhishek Saroha, Yuesong Shen, Marvin Eisenberger, Michael M. Bronstein, Daniel Cremers

Track: Extended abstract

Keywords: Transferability, Spectral Methods, Spectral Graph Theory, Proofs

TL;DR: We introduce a new approach to transferability based on information diffusion on graphs.

Abstract: This work develops a general transferability theory for graph convolutional networks; applicable to architectures based on both undirected- as well as recently introduced directed convolutional filters. Transferability is considered between graphs that are similar from the perspective of information diffusion. A detailed theoretical investigation establishes which filters render networks stable with respect to this novel approach to transferability. Illustrative examples (including graph-coarsening) showcase how newly established results may inform the design of transferable architectures in practice. Numerical experiments on real-world data validate the theoretical findings and complement the mathematical analysis.

Submission Number: 77