A mannequin that pays consideration to your graph
Welcome to the second submit about GNN architectures! Within the earlier submit, 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 submit 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 discuss in regards to the distinction between GATs and GCNs. Then let’s prepare a GAT and evaluate the accuracy with the GCN and primary neural community.
This weblog submit is a part of a sequence. Are you new to GNNs? I like to recommend you to start out with the primary submit, which explains graphs, neural networks, the dataset, and GCNs.
In my earlier submit, we noticed a GCN in motion. Let’s take it a step additional and have 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 enable the mannequin to be taught completely different significance (consideration) scores for various neighbors. They combination neighbor info through the use of consideration mechanisms (this would possibly ring a bell…