Servers Trustworthiness, Estimation, and Statistical Inference

[Submitted on 25 Apr 2024]

View a PDF of the paper titled Differentially Personal Federated Studying: Servers Trustworthiness, Estimation, and Statistical Inference, by Zhe Zhang and a pair of different authors

View PDF

Summary:Differentially non-public federated studying is essential for sustaining privateness in distributed environments. This paper investigates the challenges of high-dimensional estimation and inference underneath the constraints of differential privateness. First, we examine eventualities involving an untrusted central server, demonstrating the inherent difficulties of correct estimation in high-dimensional issues. Our findings point out that the tight minimax charges depends upon the high-dimensionality of the info even with sparsity assumptions. Second, we take into account a situation with a trusted central server and introduce a novel federated estimation algorithm tailor-made for linear regression fashions. This algorithm successfully handles the slight variations amongst fashions distributed throughout totally different machines. We additionally suggest strategies for statistical inference, together with coordinate-wise confidence intervals for particular person parameters and methods for simultaneous inference. In depth simulation experiments help our theoretical advances, underscoring the efficacy and reliability of our approaches.