Newest papers

Federated Differentially Private Optimization via Low-Rank Preconditioning

Antti Koskela, Tejas Kulkarni, Elias Raninen · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.31069 · cited by 0 works (OpenAlex) · open-access PDF

Spanish

Consideramos la optimización diferencial privada (DP) en la configuración federada, donde una estimación DP de la matriz de covarianza de características sirve como una aproximación de Hessian para modelos lineales generalizados. Los métodos centrales de DP para la estimación de covarianza tienen un coste de comunicación que crece cuadráticamente en la dimensión de datos, mientras que las líneas de base existentes con coste de comunicación lineal son solo localmente privadas y se degradan a medida que aumenta el número de clientes. Como remedio, proponemos un marco DP-PCA federado que alcanza el coste de comunicación lineal mientras proporciona garantías de DP central bajo agregación segura. Consiste en un paso de selección de rango DP, un método de potencia DP y una liberación ruidosa de la covarianza proyectada en el sub

Machine translation (Meta M2M100); the English original follows.

Abstract

We consider differentially private (DP) optimization in the federated setting, where a DP estimate of the feature covariance matrix serves as a Hessian approximation for generalized linear models. Existing central DP methods for covariance estimation have communication cost that grows quadratically in the data dimension, while existing baselines with linear communication cost are only locally private and degrade as the number of clients increases. As a remedy, we propose a federated DP-PCA framework that achieves linear communication cost while providing central DP guarantees under secure aggregation. It consists of a DP rank selection step, a DP power method, and a noisy release of the covariance projected onto the selected low-dimensional subspace. We give error bounds for this covariance estimate and for the resulting low-communication preconditioned DP gradient descent. Empirically, our proposed DP optimization method matches a centralized baseline with quadratic communication cost at a fraction of the communication.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Covariance, Hessian matrix, Covariance matrix, Computer science, Mathematical optimization, Estimation of covariance matrices

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange