Part 2 – Comparing Message Passing Based GNN Architectures


Before going through this report, I suggest the readers check out Part 1 – Intro to Graph Neural Networks with GatedGCN where we understand the motivation behind graphical neural networks (GNN), and the various problem statements that GNNs could be useful for.

The first section of this report describes the training pipeline of a message-passing based GNN. Next, we review and compare various message-passing based GNN architectures such as vanilla GCN[2], GraphSage[3], MoNet[4], GAT[5] and GatedGCN[6]. Finally, we use Sweeps by Weights & Biases to train and compare these architectures at the Node classification task.

Read the full post →

Join our mailing list to get the latest machine learning updates.