Distributed Dynamic Online Linear Regression Over Unbalanced Graphs
Songsong Cheng, Yiguang Hong · 2020
In this paper, we design a distributed scheme for dynamic online linear regression over unbalanced communication graphs. Firstly, we introduce the push-pull based method for the dynamic online linear regression problem. Besides, we remove the conventional boundness assumptions on the decision variables and gradients and establish the corresponding upper bounds instead. Furthermore, we prove that the decision variables linearly converge to the optimal solution with a certain bounded error. Finally, we provide a numerical example to illustrate the effectiveness of the proposed method.