Topic · Machine Learning & AI
← All topicsGraph Machine Learning
20 articles · 4 checkpoints · 12 deeper reads · 4 reference notes
A standalone topic: it is on no roadmap, so read it on its own terms.
Every article, in reading order
plant a flag as you finish eachRead these first
A GCN weights every neighbor by a fixed rule based on degree; a GAT instead learns which neighbors matter more, computing a different attention weight for every edge based on what the two connected nodes actually contain.
A GCN updates every node by averaging its neighbors' features, weighted so that well-connected nodes don't drown out sparsely-connected ones, the same idea as message passing, made concrete with one specific, carefully normalized weighting rule.
A graph neural network learns by having every node send a summary of itself to its neighbors, collect what its neighbors sent back, and update its own representation accordingly, repeat this a few times and each node ends up knowing about its wider neighborhood, not just itself.
A company's stock often reacts to news about its suppliers and customers days before that news is reflected in its own price, modeling the supply chain as a graph and propagating shocks across it turns a slow, indirect information channel into a tradeable signal.
Then the rest