基于分布式Agent联盟的机器学习机制  

The Study Mechanisim Based on Distributed Agent League

作　　者：刘泉[1] 乔瑞[1] 

机构地区：[1]武汉理工大学信息工程学院,武汉430070

出　　处：《武汉理工大学学报》2005年第6期79-81,共3页Journal of Wuhan University of Technology

基　　金：湖北省科技攻关项目 (2 0 0 2AA10 2B0 1)

摘　　要：提出了一种基于分布式Agent的机器学习模型。用Agent来构成机器学习的各组成部分,并由采用不同学习方法的Agent构成学习模块。为实现分布式Agent的协作学习机制,采用Agent联盟作为协作体系结构。将一种改进的遗传算法用于Agent联盟以实现分布式Agent的协作,该遗传算法控制参数动态地变化,采用粗粒度型的并行方式,一传多的交换个体方式和完全网络拓扑迁移模型。系统具有良好的性能。The paper proposed a machine-learning model based on distributed agents. The system comprised several agents, and agents with different study methods composed the learning part. To realize the cooperative study, the agent league was adopted as cooperative structure, and an improved inheritance algorithm was used in agent league to realize the cooperation of distributed agents. The heritance algorithm controlled the parameters dynamically, and adopted the collateral mode of coarse granularity. The one to multi-ones mode to exchange individuals and the all-connect-graph transference model were also adopted. The system performed very well.

关 键 词：AGENT 机器学习 协调 协作联盟 遣传算法 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]