A Distributed Online Gradient Algorithm for Multi-agent Optimization with Weight Decay

报告人：李德权 教授

单位：安徽理工大学

题目：A Distributed Online Gradient Algorithm for Multi-agent Optimization with Weight Decay

时间：2020年10月25日，下午14：00

地点：稼先楼二阶教室

摘要：In recent years, the topic of distributed optimization hasbecome a canonicalproblem in the study of networked systems. In thedistributed network, the agentscollectively minimize the globaltime-varying objective function that can be written asthe sum of several local time-varying objective functions in general. To tackle onlineconvex optimization problems in a collaborative and distributed manner, we proposed aweight decay distributed adaptive online gradient algorithm, which incorporatesdistributed optimization methods with adaptive strategies. Theoretical analysis clearlyillustrates the difference between weight decay and L_2 regularization for distributedadaptive gradient algorithms. We also analyzed the dynamic regret bound for the proposedalgorithm. Numerical experiments demonstrate that our proposed algorithm works well inpractice and compares favorably to existing distributed online optimization schemes.

报告人简介：李德权，博士、教授、硕士生导师，安徽理工大学数学与大数据学院副院长。安徽省学术和技术带头人后备人选。1996年7月本科毕业于安徽大学数理统计专业，2003年4月硕士毕业于东南大学控制理论与控制工程专业，2012年10月博士毕业于上海交通大学控制理论与控制工程专业。

主要从事分布式优化理论与方法、统计机器学习、无线传感网络分析与控制等。正主持省部级项目2项，作为核心成员正参加国家重点研发计划子课题1项；主持完成国家自然科学基金面上项目2项、省部级项目5项。获安徽省科技进步二等奖1项、安徽省自然科学三等奖2项。出版著作1部，国际期刊IEEE Access客座编辑。已在Automatica、Systems & Control Letters、International Journal of Robust and Nonlinear Control、Asian Journal of Control, Neurocomputing、Journal of SystemsScience and Complexity、Communications in Nonlinear Science and Numerical Simulation等国内外期刊和国际学术会议上发表SCI/EI检索学术论文50余篇。