GCN
Graph Convolutional Network (GCN) is a graph neural network architecture proposed by Kipf and Welling in 2017. Based on a first-order local approximation of spectral graph convolution, it propagates and aggregates node features in the spatial domain via a normalized adjacency matrix, and is one of the most influential foundational GNN models.
Timeline (last 90 days)
Unlike Transformers, which treat input as a sequence and rely on self-attention mechanisms to capture long-range dependencies, GNNs explicitly model entities and their relationships, with typical architectures including GCN, GAT, GraphSAGE, and others.
MAEM uses Simplified Graph Convolution (SGC) to build node representations; SGC is more lightweight and more stable than standard GCN, making it suitable for continual learning scenarios.
Common GNN variants include GCN (simplified based on spectral approximation), GraphSAGE (sampling and aggregation, suitable for large graphs), and GAT (introducing attention to weight neighbors).
The PS5 uses a customized AMD GPU, and its GNM API is designed with reference to AMD's GCN/RDNA architecture.
Graph Convolutional Networks (GCN) were proposed by Kipf and Welling in 2017.
Early GCNs are essentially transductive—they require seeing the complete graph structure during training and cannot handle new nodes that did not appear in the training set during inference.
The single-layer propagation formula for GCN is H^(l+1) = σ( D̃^(-1/2) à D̃^(-1/2) H^(l) W^(l) ), where à is the adjacency matrix with added self-loops and D̃ is the corresponding degree matrix.
All Facts (7)
Unlike Transformers, which treat input as a sequence and rely on self-attention mechanisms to capture long-range dependencies, GNNs explicitly model entities and their relationships, with typical architectures including GCN, GAT, GraphSAGE, and others.
50%UnverifiedMAEM uses Simplified Graph Convolution (SGC) to build node representations; SGC is more lightweight and more stable than standard GCN, making it suitable for continual learning scenarios.
50%UnverifiedCommon GNN variants include GCN (simplified based on spectral approximation), GraphSAGE (sampling and aggregation, suitable for large graphs), and GAT (introducing attention to weight neighbors).
50%UnverifiedThe PS5 uses a customized AMD GPU, and its GNM API is designed with reference to AMD's GCN/RDNA architecture.
50%UnverifiedGraph Convolutional Networks (GCN) were proposed by Kipf and Welling in 2017.
50%UnverifiedEarly GCNs are essentially transductive—they require seeing the complete graph structure during training and cannot handle new nodes that did not appear in the training set during inference.
50%UnverifiedThe single-layer propagation formula for GCN is H^(l+1) = σ( D̃^(-1/2) à D̃^(-1/2) H^(l) W^(l) ), where à is the adjacency matrix with added self-loops and D̃ is the corresponding degree matrix.
50%