GNN Tutorial Notes

0. Related url:

A Gentle Introduction to Graph Neural Networks: https://distill.pub/2021/gnn-intro/
Understanding Convolutions on Graphs: https://distill.pub/2021/understanding-gnns/

#######

supervised learning

Graph Convolutional Network

Graph Sample and Aggregate Approach
Graph Attention Network (GAT)

unsupervised learning

Graph Auto-Encoder
Graph Generative Network
Graph Spatial-Temporal Network

Basic idea:

1. Using adjacency matrix to realize matrix representation of graph structure
2. Intra-layer and inter-layer message passing through aggregation operations

#######

1.

1. What kind of data is most naturally phrased as a graph
2. What makes graphs different from other types of data
3. Walking through each of the parts of the model
4. GNN playground

A graph represents the relations (edges) between a collection of entities (nodes).

Graphs and where to find them

• Images as graphs

• Text as graphs

• Molecules as graphs

• Social networks as graphs

• Citation networks as graphs

types of problems

• graph-level
• node-level
• edge-level

The challenges of using graphs in machine learning

Graphs have up to four types of information:

• nodes
• edges
• global-context
• connectivity

Graph Neural Networks

we learn new embeddings for all graph attributes (nodes, edges, global),
but where we do not yet use the connectivity of the graph.