Reproduce “Integrating Homomorphic Encryption and Synthetic Data in FL for Privacy and Learning Quality” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Yenan Wang, Carla Fabiana Chiasserini, Elad Michael Schiller

A framework for Federated Learning, Alternating Federated Learning (Alt-FL), integrates Selective Homomorphic Encryption with an interleaved strategy of authentic and synthetic data rounds. This approach enhances model accuracy by up to 13.4% and reduces Homomorphic Encryption-related computational overhead by approximately 48% while maintaining privacy against gradient inversion attacks.

Key insights

  • Interleaving authentic and synthetic data rounds allows for a strategic reduction of Homomorphic Encryption's computational and communication burden.
  • Synthetic data not only boosts learning quality but also implicitly enhances privacy during unencrypted rounds by being statistically similar yet disjoint from authentic data, preventing effective gradient inversion attacks on real data.
  • The tunable interleaving ratio (ρ) provides a flexible mechanism to balance model accuracy, privacy protection, and resource consumption based on specific application requirements.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.