A two-stage differentially private federated estimator for high-dimensional VAR models with low-rank shared components and sparse local deviations provides non-asymptotic error bounds and consistent rank selection, outperforming single-client methods in simulations and real applications.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
stat.ME 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Private Federated Learning for High-dimensional Time Series
A two-stage differentially private federated estimator for high-dimensional VAR models with low-rank shared components and sparse local deviations provides non-asymptotic error bounds and consistent rank selection, outperforming single-client methods in simulations and real applications.