Gradient Ascent #10
graph-neural-networksknowledge-graphsgraph-theorydeep-learning
Abstraction: Overview of graph neural networks, knowledge graphs, and GNN problem settings
Key points:
- Three core GNN problem settings: node/entity classification, link prediction, and community detection
- Two main technical approaches: graph embedding (encoder-decoder) and message-passing with learned graph operators
- Knowledge graphs encode facts as (subject, predicate, object) triples; evolved from semantic web / RDF to labeled property graphs (e.g., Neo4j)
- Relational Graph Convolutional Network (R-GCN) handles multi-relational KGs by accumulating activations across relation types; implemented by Amazon DGL-KE and available via AWS Neptune ML
- Real-world complex networks (brain, protein, social) typically exhibit non-trivial topologies including scale-free degree distributions
Connections: Amazon · Aws · Graph Neural Networks · Knowledge Graphs · Graph Theory
Source: https://albertazout.substack.com/p/gradient-ascent-10