Differentially Private Coordinate Descent for Composite Empirical Risk Minimization

Machine learning models can leak information about the data used to train\nthem. To mitigate this issue, Differentially Private (DP) variants of\noptimization algorithms like Stochastic Gradient Descent (DP-SGD) have been\ndesigned to trade-off utility for privacy in Empirical Risk Minimization (ERM)\nproblems. In this paper, we propose Differentially Private proximal Coordinate\nDescent (DP-CD), a new method to solve composite DP-ERM problems. We derive\nutility guarantees through a novel theoretical analysis of inexact coordinate\ndescent. Our results show that, thanks to larger step sizes, DP-CD can exploit\nimbalance in gradient coordinates to outperform DP-SGD. We also prove new lower\nbounds for composite DP-ERM under coordinate-wise regularity assumptions, that\nare nearly matched by DP-CD. For practical implementations, we propose to clip\ngradients using coordinate-wise thresholds that emerge from our theory,\navoiding costly hyperparameter tuning. Experiments on real and synthetic data\nsupport our results, and show that DP-CD compares favorably with DP-SGD.\n

Paper

Similar papers

© 2026 NYSGPT2525 LLC