New federated learning method enhances differential privacy on complex manifolds

Researchers have developed DP-RFedProj, a novel method for federated learning that enhances differential privacy on compact submanifolds. This approach privatizes gradients in a fixed ambient geometry, allowing clients to perform multiple local steps and participate intermittently while maintaining an explicit record-level f-DP guarantee. The method establishes convergence for Riemannian SGD and projected averaging, with privacy costs dependent on the manifold's intrinsic dimension. Experiments on federated PCA and EEG motor-imagery classification demonstrated superior utility compared to Euclidean baselines, requiring significantly less noise for the same privacy budget. AI

IMPACT Enhances privacy guarantees for federated learning on complex data structures, potentially enabling more secure collaborative model training.

RANK_REASON The cluster contains an academic paper detailing a new method for federated learning with differential privacy. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv stat.ML →

• DP-RFedProj
• Grassmannian
• Riemannian manifold
• Riemannian SGD
• SPDnet
• Stiefel manifold

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →