A mannequin that pays consideration to your graph
Welcome to the second put up about GNN architectures! Within the earlier put up, we noticed a staggering enchancment in accuracy on the Cora dataset by incorporating the graph construction within the mannequin utilizing a Graph Convolutional Community (GCN). This put up explains Graph Consideration Networks (GATs), one other elementary structure of graph neural networks. Can we enhance the accuracy even additional with a GAT?
First, let’s speak concerning the distinction between GATs and GCNs. Then let’s practice a GAT and examine the accuracy with the GCN and fundamental neural community.
This weblog put up is a part of a sequence. Are you new to GNNs? I like to recommend you to begin with the primary put up, which explains graphs, neural networks, the dataset, and GCNs.
In my earlier put up, we noticed a GCN in motion. Let’s take it a step additional and take a look at Graph Consideration Networks (GATs). As you would possibly bear in mind, GCNs deal with all neighbors equally. For GATs, that is completely different. GATs permit the mannequin to study completely different significance (consideration) scores for various neighbors. They combination neighbor data by utilizing consideration mechanisms (this would possibly ring a bell…
