Natural Graph Networks [article]

Pim de Haan, Taco Cohen, Max Welling
2020 arXiv   pre-print
A key requirement for graph neural networks is that they must process a graph in a way that does not depend on how the graph is described. Traditionally this has been taken to mean that a graph network must be equivariant to node permutations. Here we show that instead of equivariance, the more general concept of naturality is sufficient for a graph network to be well-defined, opening up a larger class of graph networks. We define global and local natural graph networks, the latter of which are
more » ... as scalable as conventional message passing graph neural networks while being more flexible. We give one practical instantiation of a natural network on graphs which uses an equivariant message network parameterization, yielding good performance on several benchmarks.
arXiv:2007.08349v2 fatcat:bble7vlwiza77k7i2jv3tvvcce