Federated GNN for Distributed Link Prediction
Memory-efficient federated graph convolutional network training on commodity hardware (IEEE Big Data 2020)
Federated learning for collaborative, privacy-preserving graph learning across organizations that run distributed graph databases. Built on the JasmineGraph distributed graph database.
- Built memory-efficient distributed and federated GCN training that handles very large graphs on commodity hardware, using JasmineGraph and parallel workers.
- Designed aggregation mechanisms for heterogeneous multi-organization graph learning that keep each organization’s data private.
- Peer-reviewed publication at IEEE Big Data 2020.
Technologies: C++, Python, StellarGraph, TensorFlow
Supervisors: Prof. Sanath Jayasena, Dr. Miyuru Dayarathna
| Links: Publication | Source |