A decentralized Nesterov gradient method for stochastic optimization over unbalanced directed networks

Abstract

Decentralized stochastic gradient methods play significant roles in large‐scale optimization that finds many practical applications in machine learning and coordinated control. This paper studies optimization problems over unbalanced directed networks, where the mutual goal of agents in the network is to optimize a global objective function expressed as a sum of local objective functions. Each agent using only local computation and communication in the networks is assumed to get access to a stochastic first‐order oracle. In order to devise a noise‐tolerant decentralized algorithm with accelerated linear convergence, a decentralized Nesterov gradient algorithm with the constant step‐size and parameter using stochastic gradients is proposed in this paper. The proposed algorithm employing a gradient‐tracking technique is proved to converge linearly to an error ball around the optimal solution via the analysis on a linear system when the positive constant step‐size and parameter are sufficiently small. We further recover the exact linear convergence for the proposed algorithm with exact gradients under the same selection conditions of the constant step‐size and parameter. Some real‐world data sets are used in simulations to validate the correctness of the theoretical findings and practicability of the proposed algorithm.

Mobility‐aware optimal trade planning and mean field game‐based decentralized charging control for large‐scale electric vehicles

“…There was research focusing on the application of decentralized controller into the large-scale web-winding system [10]. On the other hand, the decentralized controller is an effective tool in dealing with the system with random disturbances, such as in the network area [11,12]. In decentralized control, the mean field game (MFG) is an alternative route when encountering large-scale agents, which was introduced by Lasry and Lions and Huang et al in 2007 to manage a large number of agents in a non-cooperative game problem [20,21].…”

Multi‐AGV route planning in automated warehouse system based on shortest‐time Q‐learning algorithm

“…Common multi‐AGV control systems can be classified into centralized systems and distributed systems. We choose to use the decentralized architecture as described in earlier studies [30, 31], as shown in Figure 1. The upper layer consists of a database that stores the necessary information for the environment and scheduling, a scheduling module, route planning module, and a supervisory control module.…”

A maximum principle for progressive optimal control of mean‐field forward–backward stochastic system involving random jumps and impulse controls

“…Nowadays, mean-field control and game problems have attracted and gained more and more researchers' attentions, because of the extensive applications and significant theoretical values in many fields, such as economics, information technology, engineering, and system control [1][2][3][4][5][6][7][8][9][10]. There are two typically used problem frames and models for the mean-field control.…”