English

Smoothed Normalization for Efficient Distributed Private Optimization

Machine Learning 2025-02-20 v1 Cryptography and Security Distributed, Parallel, and Cluster Computing Optimization and Control Machine Learning

Abstract

Federated learning enables training machine learning models while preserving the privacy of participants. Surprisingly, there is no differentially private distributed method for smooth, non-convex optimization problems. The reason is that standard privacy techniques require bounding the participants' contributions, usually enforced via clipping\textit{clipping} of the updates. Existing literature typically ignores the effect of clipping by assuming the boundedness of gradient norms or analyzes distributed algorithms with clipping but ignores DP constraints. In this work, we study an alternative approach via smoothed normalization\textit{smoothed normalization} of the updates motivated by its favorable performance in the single-node setting. By integrating smoothed normalization with an error-feedback mechanism, we design a new distributed algorithm α\alpha-NormEC\sf NormEC. We prove that our method achieves a superior convergence rate over prior works. By extending α\alpha-NormEC\sf NormEC to the DP setting, we obtain the first differentially private distributed optimization algorithm with provable convergence guarantees. Finally, our empirical results from neural network training indicate robust convergence of α\alpha-NormEC\sf NormEC across different parameter settings.

Keywords

Cite

@article{arxiv.2502.13482,
  title  = {Smoothed Normalization for Efficient Distributed Private Optimization},
  author = {Egor Shulgin and Sarit Khirirat and Peter Richtárik},
  journal= {arXiv preprint arXiv:2502.13482},
  year   = {2025}
}

Comments

36 pages

R2 v1 2026-06-28T21:49:42.233Z