GraphSAGE

Inductive Representation Learning on Large Graphs

William L. Hamilton, Rex Ying and Jure Leskovec · NIPS 2017

Read paper · arXiv:1706.02216

Aggregate sampled neighborhoods with shared feature transformations and transfer the encoder to an unseen graph.

What you will implement

Algorithm 1 mean encoding, per-hop normalization and gradients; 100 actual synthetic training updates and evaluation on a distinct graph, without reproducing the full TensorFlow configuration or paper benchmarks.

• 100. GraphSAGE: Sampled Neighborhood Means and Inductive Encoding

  Aggregate sampled neighborhood features and generate embeddings for unseen nodes with shared weights.

  Implement masked means and Algorithm 1 synchronous sampled hops; sampling tables, validation and small inductive classification training are supplied.