Graph SAGE - Inductive Representation Learning on Large Graphs - GNN Paper Explained

Dive deep into the Graph SAGE paper, exploring the groundbreaking approach for using Graph Neural Networks (GNNs) on large-scale graphs. Learn about the key components of Graph SAGE, including its training process, neighborhood functions, and aggregator functions. Understand the method's expressiveness, mini-batch implementation, and how it addresses problems with previous graph embedding techniques. Compare Graph SAGE to other popular GNN architectures like GCN and GAT, gaining insights into its advantages and applications in processing large-scale graph data.

Syllabus

Intro
Problems with previous methods
High-level overview of the method
Some notes on the related work
Pseudo-code explanation
How do we train Graph SAGE?
Note on the neighborhood function
Aggregator functions
Results
Expressiveness of Graph SAGE
Mini-batch version
Problems with graph embedding methods drift
Comparison with GCN and GAT