Architectures · advanced · concept 71 of 176
Graph Neural Networks (GNNs)
Networks that operate on graphs by passing messages between connected nodes, so a node's representation absorbs its neighborhood. Where CNNs assume a grid and transformers assume a sequence, GNNs assume relationships, which is what fraud rings, molecules, and social networks actually are.
Key terms
Message passingNode embeddingsGraph convolutionAggregationLink prediction
Learn these first
Where you meet it in the real world
Fraud detection, drug discovery, recommendation graphs, traffic prediction, chip placement
Videos
▶ Stanford CS224W: Machine Learning w/ Graphs I 2023 I Graph Neural Networks ↗
Stanford Online · YouTube
▶ Graph Neural Networks Explained: A Clear Guide to GNN Basics & Models ↗
IBM Technology · YouTube
Guides and articles
Courses, papers, and more