Federated GNN for Distributed Link Prediction

Memory-efficient federated graph convolutional network training on commodity hardware (IEEE Big Data 2020)

Applied federated learning for facilitating collaborative, privacy-preserving graph learning among organizations that use distributed graph database systems. Implementation was based on JasmineGraph distributed graph database system.

  • Built memory-efficient distributed/federated GCN training enabling training on very large graphs on commodity hardware using JasmineGraph and parallel workers.
  • Developed novel aggregation mechanisms for heterogeneous multi-organization graph learning while maintaining privacy.
  • Peer-reviewed publication at IEEE Big Data 2020.

Technologies: C++, Python, StellarGraph, TensorFlow

Supervisors: Prof. Sanath Jayasena, Dr. Miyuru Dayarathna

Links: Publication Source