Quantized Primal-Dual Algorithms for Network Optimization With Linear Convergence

• 成果类型: Article
• 署名作者: Chen, Ziqin; Liang, Shu; Li, Li; Cheng, Shuming
• 署名单位: Tongji University
• 刊物名称: IEEE TRANSACTIONS ON AUTOMATIC CONTROL
• ISSN/ISSBN: 0018-9286
• DOI: 10.1109/TAC.2023.3266018
• 发表日期: 2024
• 页码: 471-478
• 关键词: convergence; Quantization (signal); bandwidth; Artificial neural networks; estimation; Convex functions; Prediction algorithms; distributed convex optimization; linear convergence rate; primal-dual algorithm; quantized communication

摘要

This note investigates a network optimization problem in which a group of agents cooperate to minimize a global function under the practical constraint of finite-bandwidth communication. We propose an adaptive encoding-decoding scheme to handle the quantization communication between agents. Based on this scheme, we develop a continuous-time quantized distributed primal-dual algorithm for the network optimization problem. Our algorithm achieves linear convergence to an exact optimal solution. Furthermore, we obtain the relationship between the communication bandwidth and the convergence rate. Finally, we use a distributed logistic regression problem to illustrate the effectiveness of our methods.