T
N
Last name: TREMBLAY
First name: Nicolas
No description
International audience
We study the convergence of message passing graph neural networkson random graph models to their continuous counterpart as the number ofnodes tends to infinity. Until now, this convergence was only known forarchitectures with aggregation functions in the form of normalized means, or,equivalently, of an application of classical operators like...
We study the limit behavior of GNNs on large random graphs. We consider GNNs with a generic aggregation function and show that under mild regularity conditions, they converge to a "continuous" counterpart. We provide some non asymptotic bounds with high probability for this convergence which encompass several cases of aggregation such as, the...
We study the convergence of message passing graph neural networkson random graph models to their continuous counterpart as the number ofnodes tends to infinity. Until now, this convergence was only known forarchitectures with aggregation functions in the form of normalized means, or,equivalently, of an application of classical operators like...
We explore the expressive power of Graph Neural Networks (GNNs) in the context of large random graphs. In this document, we focus on Spectral GNNs, defined over random graphs drawn from a latent space Random Graph Model (RGM). Definition 1 (GNN). Let G be a graph with adjacency matrix A. A GNN Φ G over G, is a nonlinear map from the set of...