|
| 分布式流言push-sum无梯度算法 |
| Distributed gossip-based push-sum gradient-free algorithm |
| 投稿时间：2017-07-26 |
| DOI： |
| 中文关键词: 多个体网络 网络优化 分布式优化 流言算法 push-sum算法 无梯度算法 |
| 英文关键词: multi-agent network network optimization distributed optimization gossip algorithm push-sum algorithm gradient-free algorithm |
| 基金项目:国家自然科学基金资助项目(61472003)；高校学科(专业)拔尖人才学术资助重点项目(gxbjZD2016049)；安徽省学术和技术带头人及后备人选科研活动经费资助项目(2016H076). |
|
| 摘要点击次数: 4525 |
| 全文下载次数: 3196 |
| 中文摘要: |
| 研究多个体网络中所有个体目标函数之和最小值问题,其中每个个体仅知其自身目标函数且仅可与其邻居个体交互信息。鉴于个体目标函数通常非光滑,同时个体间单变量信息通信有一定局限性,本文提出一种分布式流言push-sum无梯度算法求解此优化问题。假设每个个体都具有一个服从泊松分布的控制时钟,时钟的每次转动表示随机选择的个体之间进行信息更新。进一步地,在网络连通条件下证明了所提算法的收敛性。数值仿真结果表明,与现有的分布式流言无梯度优化算法相比,本文算法具有更快的收敛速度。 |
| 英文摘要: |
| This paper studies how to collaboratively minimize the whole objective function of the multi-agent network, wherein each agent just knows its own objective function and only interacts with its neighboring agents. Because each agent’s local objective function is usually non-smooth and the method of single variable information communication among agents has some limitations, a distributed gossip-based push-sum gradient-free (DGPSGF) algorithm is proposed to solve this optimization problem. It is assumed that each agent has a local Poisson clock, and at each tick of its clock rotation, the agent interacts with randomly selected agents. Furthermore, the convergence of DGPSGF algorithm is proved under the mild assumption on the network connectivity. Numerical simulation results show that, compared with the existing distributed gossip-based gradient-free algorithm, the proposed DGPSGF algorithm has faster convergence rate. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
