We present an approach for analyzing message passing graph neural networks
(MPNNs) based on an extension of graphon analysis to a so called graphon-signal
analysis. A MPNN is a function that takes a graph and a signal on the graph (a
graph-signal) and returns some value. Since the inpu