摘要
在大脑的神经网络发育过程中有大量的突触产生,然后随着时间的推移逐渐修剪。这种策略在设计工程网络结构时并不常用,因为添加一个即将被删除的连接是一种不必要的资源浪费。对于大型的分布式路由网络,超连通后的主动修剪显著增强了网络功能。而在优化网络结构方面,全局剪枝率起着至关重要的作用。通过对一个计算路由网络模型进行理论分析和仿真实验得出结论,降低速率可以使网络更加健壮和高效。
During the development of the brain's neural network,a large number of synapses are produced and gradually pruned over time.This strategy is not commonly used when designing an engineering network structure because adding a connection that is about to be deleted is an unnecessary waste of resources.For large distributed routing networks,hyperconnected active pruning significantly enhances network functions,and global pruning rate plays an important role in optimizing network structure.Through the theoretical analysis and simulation of a computational routing network model,it is concluded that reducing the rate can make the network more robust and efficient.
作者
周宗玄
徐慧
Zhou Zongxuan;Xu Hui(Wen Zheng College of Soochow University,Suzhou Jiangsu 215000,China)
出处
《信息与电脑》
2018年第17期135-136,共2页
Information & Computer
基金
江苏省大学生创新创业训练计划项目(项目编号:201713983009Y)
关键词
减速修剪
分布式网络
增长算法
decreasing-rate pruning
distributed network
growing algorithm
作者简介
周宗玄(1997-),男,江苏苏州人,本科。研究方向:计算机科学与技术。;徐慧(1996-),女,四川内江人,本科。研究方向:通信工程。
