三元XOR问题的神经网络学习  

Neural Networks Training On the Trinary XOR

作　　者：赵娣[1] 田大钢[1] 

机构地区：[1]上海理工大学管理学院

出　　处：《数学理论与应用》2015年第3期64-69,共6页Mathematical Theory and Applications

基　　金：沪江基金(A14006);上海市一流学科项目(S1201YLXK)资助

摘　　要：一个给定的神经网络能否学习一个给定的样本集一直是一个有趣的问题.对于单隐层前馈神经网络,当隐层神经元的数量和样本的数量相等时,这个问题是平凡的.而对于隐层神经元的数量少于样本数量这种情况,相关讨论还很少,值得更多的关注和研究.本文关于这个问题提出了一种方法,并将其应用到三元XOR问题中,给出了一些初步结果.Whether a given neural network can learn a given sample set is always an interesting problem. For single hidden layer feedforward neural networks, when the number of hidden layer neurons is equal to the number of sampies, the problem has been discussed sufficiently. For the case where the number of hidden layer neurons is less than the number of samples, there is very little discussion and it is worthy to be referred. In this paper, a general way to consider the problem is proposed, and some results are given for the trinary XOR.

关 键 词：前馈型神经网络 样本学习 三元XOR 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]