Problem

Has GraphConvolutionalLayer but missing major GNN architectures.

Existing

• src/NeuralNetworks/Layers/GraphConvolutionalLayer.cs
• src/RetrievalAugmentedGeneration/KnowledgeGraph.cs (graph storage)

Missing Implementations

Core GNN Architectures (CRITICAL):

• GAT (Graph Attention Networks)
• GraphSAGE (inductive learning)
• GIN (Graph Isomorphism Network)

Message Passing (HIGH):

• General Message Passing Neural Networks
• Edge-conditioned convolutions
• Heterogeneous graph support

Advanced (MEDIUM):

• Graph Transformers
• Principal Neighbourhood Aggregation
• Directional Graph Networks

Tasks (HIGH):

• Node classification
• Link prediction
• Graph classification
• Graph generation

Use Cases

• Social network analysis
• Molecular property prediction
• Knowledge graph reasoning
• Recommendation systems

Architecture

• Expand src/NeuralNetworks/Layers/Graph/
• Interface: IGraphConvolutionLayer
• Graph data loaders

Success Criteria

• OGB (Open Graph Benchmark) datasets
• Benchmarks on Cora, CiteSeer, PubMed
• Molecular datasets (ZINC, QM9)