SUBGRADIENT PROJECTION OVER DIRECTED GRAPHS USING SURPLUS CONSENSUS

摘要

In this paper, we propose Directed-Distributed Projected Sub-gradient (D-DPS) to solve a distributed constrained optimization problem over a sensor network. Sensors collaboratively minimize a sum of convex functions, which are only locally known and constrained to some commonly known convex set. D-DPS is based on surplus consensus, which overcomes the information asymmetry caused by directed communication and has a convergence rate of $Oleft ({ {frac {ln k}{sqrt {k} }} }right )$.