At the core of every Graph Neural Network is a simple idea: a node updates its representation by combining its own features with information from its neighbors. This process is called message passing. Each node receives “messages” from adjacent nodes, aggregates them using a permutation-invariant function such as sum, mean, or max, and then applies a learnable transformation. Formally, a node’s representation at layer l+1depends on the representations of its neighbors at layer l. Stacking layers expands the receptive field from 1-hop neighbors to multi-hop neighborhoods. Message passing is the mechanism that allows structure to shape representation.
