Graph Neural Network

• model that takes the node embeddings X and the adjacency matrix A as inputs and passes them through a series of K layers.
• The node embeddings are updated at each layer to create intermediate “hidden” representations $H_k$ before finally computing output embeddings $H_K$
• At the start of this network, each column of the input node embeddings X just contains information about the node itself. At the end, each column of the model output $H_K$ includes information about the node and its context within the graph.
• similar to word Embedding for Transformer
• 

Graph Level Tasks

Node Level Tasks

Edge Prediction Tasks