哈尔滨工业大学（深圳）学术讲座

演讲人Speaker： 宋国辉  

题目Title: Decentralized Algorithms for Spatially Distributed Systems

时间Date：  2023 年 6月29日       Time：11：00-12：00

地点Venue：腾讯会议室 489 428 894

内容摘要Abstract：  

  In a centralized system, all processing and decision-making is handled by a single entity, which can become a bottleneck as the system grows in size and complexity. Decentralized algorithms distribute the processing load across multiple nodes, allowing the system to scale much more effectively. Many decentralized  algorithms distribute the global objective function (usually the sum of many local objective functions) across multiple nodes such that each node only handle its own local objective function. The state variable is ``copied" to each node and communication between neighboring nodes helps them reach a consensus in the convergence of iterations. However, the computational cost for each agent can still be high if the common state variable has a large dimension. To address this, we would further divide the global state variable into multiple local state variables, so that each node only handles a few components of the global state variable. In particular, we will analyze its performance on spatially distributed systems.

个人简介(About the speaker):

宋国辉，美国 Old Dominion University 数学与统计系副教授。主要从事机器学习和优化算法的研究。研究领域包括再生核空间的理论基础，傅立叶数据分析，和图优化问题的分布算法。相关研究成果发表在Applied and Computational Harmonic Analysis，SIAM Journal on Numerical Analysis， SIAM Journal on Scientific Computing，SIAM Journal on Imaging Sciences等权威期刊。