Distributed ADMM With Synergetic Communication and Computation

In this article, we propose a novel distributed alternating direction method of multipliers (ADMM) algorithm with synergetic communication and computation, called SCCD-ADMM, to reduce the total computation cost system. Explicitly, in proposed algorithm, each node interacts only part its neighboring nodes, number which is progressively determined according heuristic searching procedure, takes into account both predicted convergence rate costs at iteration, resulting trade-off between computation. Then chooses nodes an importance sampling distribution derived theoretically minimize variance latest information it locally stores. Finally, updates local new update rule adapts nodes. We prove provide upper bound brought by randomness. Extensive simulations validate excellent performances terms variance, overall cost, impact network topology as well time for evaluation, comparison traditional counterparts.