Abstract

In this talk, I will discuss some ingredients to better understand the behavior of graph machine learning, and especially GNNs, on large random graphs. I will present the random geometric graph model from the probability & statistics community, and how we can draw some conclusions regarding the convergence and the stability of some deep architectures. Based on joint works w/ A. Bietti, M. Cordonnier, N. Keriven, N. Tremblay