神经网络系统中的数据一致性算法  

A novel data consistency algorithm based on neural networks

作　　者：薛飞[1] 宋之文[1] 王征[1] 

机构地区：[1]西南财经大学经济信息工程学院与金融研究中心,四川成都610074

出　　处：《武汉大学学报（工学版）》2010年第5期662-666,共5页Engineering Journal of Wuhan University

基　　金：四川应用基础研究项目(编号:2008JY0070-2);科技型中小企业技术创新基金项目(编号:04C26225110223);西南财经大学金融智能与金融工程四川省重点实验室公开项目(编号:FIFE2010-P05).

摘　　要：数据处理的步调一致是神经网络系统的重要问题.根据神经网络的特点,提出了新型的数据一致性算法.该算法的运行范围限制在请求神经元到根神经元之间,采用回溯方法寻找仲裁神经元,并采用逻辑时戳保证消息的时序性;算法采用残存树探测方法应对网络改造等问题.性能分析与仿真证明,该算法具有较低的消息复杂度、较短的响应延迟以及较好的容错性能.Data processing consistency is an important problem of neural network systems. According to the properties of neural network systems, a novel algorithm was presented for then：. Based on the executing sets between the root neuron and the requestor neurons, the algorithm searched a decision neuron by trace methods. And logical timestamps were utilized to guarantee the time sequence. Furthermore, the max remainder probe methods were employed to deal with some network improvement. Analysis and simu- lation results show that it has lower message complexities; shorter response delay and better fairness than the traditional algorithms do so.

关 键 词：神经网络 数据一致性 回溯 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]