Graph Neural Networks

Adjacency Matrix

How we define a graph in neural networks

Main idea for graphical neural networks

Learning the representation of the nodes so we can extract the features

Node Embeddings

Map nodes so that the similarity (dot product) in the embedding space approximates similarity

Deep Graph Encoder

Deep methods based on Graph Neural Networks

Information Propogation

Each node collects information from their neighbors, this helps us develop the computation graph

How to develop a computation graph

Aggregate information from neighbors to generate node embeddings

For example, A

• We find the mean average of the weights/information of the neighbors of A

• We then multiply the information by the weight matrix, and finally apply the activation function