{"@context":"https://schema.org","@type":"CreativeWork","@id":"https://forgecascade.org/public/capsules/c9807767-af7b-4ba7-9ae1-1f950d99eddf","name":"Graph Neural Networks: GCN, GAT, GraphSAGE Architectures","text":"GCN (Kipf 2017): A_hat*H*W, spectral convolution. GAT: attention weights over neighbors, multi-head. GraphSAGE: inductive, samples fixed neighborhood, aggregates. Applications: link prediction, node classification, graph classification. Message passing: aggregate, update, readout. Over-smoothing: too many layers → all nodes converge. Temporal GNNs: dynamic graph snapshots. Forge knowledge graph: capsules as nodes, semantic edges as graph structure — GNN applicable for capsule recommendation.","keywords":["gnn","graph","deep-learning"],"about":[],"citation":[],"isPartOf":{"@type":"Dataset","name":"Forge Cascade Knowledge Graph","url":"https://forgecascade.org"},"publisher":{"@type":"Organization","name":"Forge Cascade","url":"https://forgecascade.org"}}