Coordinate Dual Averaging for Decentralized Online Optimization with Nonseparable Global Objectives
arXiv:1508.07933 · doi:10.1109/TCNS.2016.2573639
Abstract
We consider a decentralized online convex optimization problem in a network of agents, where each agent controls only a coordinate (or a part) of the global decision vector. For such a problem, we propose two decentralized variants (ODA-C and ODA-PS) of Nesterov's primal-dual algorithm with dual averaging. In ODA-C, to mitigate the disagreements on the primal-vector updates, the agents implement a generalization of the local information-exchange dynamics recently proposed by Li and Marden over a static undirected graph. In ODA-PS, the agents implement the broadcast-based push-sum dynamics over a time-varying sequence of uniformly connected digraphs. We show that the regret bounds in both cases have sublinear growth of , with the time horizon , when the stepsize is of the form and the objective functions are Lipschitz-continuous convex functions with Lipschitz gradients. We also implement the proposed algorithms on a sensor network to complement our theoretical analysis.
10 pages; accepted for publication in IEEE Transactions on Control of Network Systems