Quantized Communication-Based Distributed Optimization over Time-Varying Directed Networks

Recent advances in networked control and distributed systems require the development of scalable algorithms that consider the decentralized characteristic of the problem and communication restrictions. The interest in solving distributed consensus optimization problems of multi-agent systems has been growing. Distributed consensus optimization problems are modeled as minimizing a global objective function by multiple agents over a network. The formulation of distributed consensus optimization has been paid extensive attention due to its widespread applications, e.g., large-scale machine learning [1, 2], model predictive control [3], cognitive networks [4, 5], source localization [6, 7], resource allocation or scheduling [8], message routing [9], distributed spectrum sensing [5], statistical inference and learning [6, 10–12]. In these applications, without need of putting all the parameters together which define the optimization problem, decentralized nodes that only have a local subcollection of such parameters, collaboratively achieve the minimizer of the global objective function. In light of commonly used distributed optimization algorithms, by introducing a large number of nodes with a certain ability of calculation and communication, a complex global optimization problem is spit and distributed to those nodes. Through locally calculating and communicating with neighboring nodes, global optimal solutions can be obtained.