Graph Neural Networks

GNN Fundamentals

Neural networks for graph data.

Graph Representation

Nodes: entities. Edges: relationships. Node features: attributes. Adjacency matrix: connectivity.

Message Passing

Aggregate neighbor information. Update node representation. Repeat for multiple layers.

Types

GCN: graph convolution. GraphSAGE: sampling neighbors. GAT: attention over neighbors.

Applications

Molecular property prediction. Social network analysis. Knowledge graphs.

Key Takeaways

1. GNNs handle graph-structured data
2. Message passing aggregates neighbor info
3. Used for molecules, social networks