GRAPH NEURAL NETWORKS · A FIELD GUIDE A tiny grid. A local rule. A wider view. Follow a feature from one bus to the next. Learn how messages become representations—and what the graph alone cannot promise.
Start reading → Five-node message graph Undirected toy graph: AB, AC, BC, CD gate 1, DE. A state 2.000; B state 0.000; C state 1.000; D state 3.000; E state 0.000. Layout is not geography; values are dimensionless. C–D: 1 gate A 2.000 B 0.000 C 1.000 D 3.000 E 0.000 Five invented node features. Every number in this graph can be traced by hand.
For the applied-ML reader. We introduce graph notation, shared functions and permutation as they become useful. No electrical-engineering background or GNN framework required.
01 A grid, before the network Five buses. Five connections. One question: what can each node learn from its neighbors?
↗ 02 One round of messages Freeze the old states. Gather neighbor information. Apply the same update everywhere.
↗ 03 How far can a change travel? Connected does not mean immediately visible. Each local layer extends the dependency path.
↗ 04 Same graph, different order A node’s answer should follow its identity, not its row number.
↗ 05 When mixing stops helping More context is useful only if the representation keeps the distinctions the task needs.
↗