Reviews: Mean-field theory of graph neural networks in graph partitioning

A GNN (graph neural network) is a neural network whose input is a graph. This paper studies the problem of using a GNN to detect clusters in a graph drawn from the 2-groups SBM (stochastic block model: a popular model for random graphs with community structure). Although it is already known how to optimally solve the SBM (i.e. Thus, it is of interest to analyze the performance of GNN on the SBM. In the GNN architecture studied here, each layer has a node for each vertex in the graph.