基于跨越连接的多层前馈神经网络结构分析  

Research on structure of multi-layer feed-forward neural network with cross connections

作　　者：唐云岚[1,2] 高妍方[1] 谭旭[1] 陈英武[1] 

机构地区：[1]国防科学技术大学信息系统与管理学院,长沙410073 [2]武警工程学院通信工程系,西安710086

出　　处：《计算机工程与应用》2009年第32期45-47,共3页Computer Engineering and Applications

基　　金：国家自然科学基金No.70272002;高等院校博士学科点专项科研基金No.20059998019~~

摘　　要：任何连接方式的神经网络总可以归结为跨越连接网络。在传统多层前馈神经网络算法的基础上,提出了完全全连接神经网络的概念,给出了基于跨越连接的多层前馈神经网络算法。通过分析多层前馈神经网络的误差函数,从理论上证明了:相对于无跨越连接网络,基于跨越连接的多层前馈神经网络能以更加简洁的结构逼近理想状态。最后,用一个隐层神经元解决了XOR问题。Neural networks with any kind of connections can always be sorted as cross-connected ones.According to traditional multi-layer feed-forward neural network,this paper elaborates the concept of completely-fully connected neural network and then puts forward a cross-connected multi-layer feed-forward neural network algorithm.By analyzing the error function of multi-layer feed-forward neural network,it can be theoretical proved that the cross-connected neural network can reach ideal results with more concise framework compared with the non-cross connected neural network.Lastly,the proposed algorithm is tested on the XOR problem,which is well solved by using only one hidden neuron.

关 键 词：跨越连接 多层前馈神经网络 隐层结构 XOR问题 

分 类 号：TP13[自动化与计算机技术—控制理论与控制工程]