1: \begin{abstract}

2: This paper studies distributed algorithms for (strongly convex) composite optimization problems  over mesh networks, subject to quantized communications.

3: Instead of focusing on a specific algorithmic design, a black-box model is proposed, casting linearly convergent distributed algorithms in the form of  fixed-point iterates.

4: The algorithmic model is equipped  with a novel random or deterministic

5:  \emph{Biased Compression} (BC) rule    on the quantizer design, and

6:   a new \emph{Adaptive encoding Non-uniform Quantizer} (ANQ)

7:  coupled with a communication-efficient encoding scheme,

8:  which  implements the BC-rule using a {\it finite} number of bits (below machine precision).  

9:     This fills a gap existing in most state-of-the-art

10:   quantization schemes, such as those based on the popular  compression rule, which rely on communication of some scalar signals

11:   with {negligible quantization} error (in practice quantized at the machine precision).

12:  A unified communication  complexity analysis is developed  for the black-box model, determining   the average number of bits required to reach a solution of the optimization problem within a target accuracy. It is shown that the proposed BC-rule preserves   {\it linear} convergence of the unquantized algorithms, and a trade-off between convergence rate and communication cost under {ANQ-based} quantization is characterized.

13: Numerical results validate our theoretical findings and show that distributed algorithms equipped with the proposed ANQ have more favorable communication cost than algorithms using state-of-the-art quantization rules.

14: \end{abstract}

15: