Linear convergence of random dual coordinate incremental aggregated gradient methods

Abstract

In this paper, we consider the dual formulation of minimizing $\sum_{i\in I}f_i(x_i)+\sum_{j\in J} g_j(\mathcal{A}_jx)$ with the index sets $I$ and $J$ being large. To address the difficulties from the high dimension of the variable $x$ (i.e., $I$ is large) and the large number of component functions $g_j$ (i.e., $J$ is large), we propose a hybrid method called the random dual coordinate incremental aggregated gradient method by blending the random dual block coordinate descent method and the proximal incremental aggregated gradient method. To the best of our knowledge, no research is done to address the two difficulties simultaneously in this way. Based on a newly established descent-type lemma, we show that linear convergence of the classical proximal gradient method under error bound conditions could be kept even one uses delayed gradient information and randomly updates coordinate blocks. Three application examples are presented to demonstrate the prospect of the proposed method.