Accelerated online averaging over networks

Online learning has been extensively investigated due to the rapidly growing demand for capabilities to process the massive data collected sequentially. Most of the existing works have studied online problems over the centralized framework. In this paper, we mainly focus on the decentralized online averaging 问题 and propose a novel accelerated algorithm. We also extend our approach to the decentralized online least squares regression. We establish both algorithmic convergence theories and statistical guarantees to demonstrate the acceleration of our algorithms. Moreover, we are the first to provide the central limit 定理 of the decentralized online least squares estimator. Several numerical simulations are conducted to demonstrate the feasibility and effectiveness of our proposed algorithms.