Fairness-Aware Federated Graph Learning

Addressing fairness and bias in federated learning settings over graph-structured data.

Venue: EAAMO 2025

FairWAG addresses fairness challenges in federated graph learning, where multiple parties collaboratively train models without sharing raw data. This work proposes a weighted aggregation strategy that balances model performance with fairness across heterogeneous clients, ensuring that the globally aggregated model does not disproportionately disadvantage certain groups or subpopulations.

FairWAG framework overview

Key contributions:

  • Fairness-aware aggregation strategies for federated graph neural networks
  • Theoretical analysis of fairness-utility tradeoffs in federated settings
  • Experiments on real-world social and citation networks