Graph Neural Networks are neural networks specifically adapted to graph structures by exchanging information between nodes and their neighbors.
The basic idea is that each node updates its state by aggregating features of neighboring nodes and edges.
Typical GNN models use iterative message passing and aggregation steps to capture local and global graph information.
This technique is applied in areas such as molecular analysis, recommendation systems, and network predictions.
Various architectures like Graph Convolutional Networks (GCN) or Graph Attention Networks (GAT) extend the concept through different aggregation and weighting methods.