1: \begin{abstract}  

2: \noindent \textbf{Abstract:} Federated learning has shown its advances over the last few years but is facing many challenges, such as how algorithms save communication resources, how they reduce computational costs, and whether they converge. To address these issues, this paper proposes exact and inexact ADMM-based federated learning. They  are not only communication-efficient but also converge linearly under very mild conditions, such as convexity-free and irrelevance to data distributions. Moreover,  the inexact version has low computational complexity, thereby alleviating the computational burdens significantly.

3: \vspace{0.3cm}

4:  

5: \noindent{\bf \textbf{Keywords}:}  ADMM-based federated learning, efficient communications, low computational costs, global convergence with linear rate 

6:  

7: %\vspace{3mm}

8: %\noindent{\bf \textbf{Mathematical Subject Classification}:} 90C27 $\cdot$ 90C30 $\cdot$ 90C26 $\cdot$ 90C46

9: \end{abstract}

10: