摘要
本文首先将Ｋｏｈｏｎｅｎ网络的训练过程划分成两个阶段，即自组织拓扑保持映射形成阶段和训练阶段，进一步分析了后训练阶段的拓扑保持特性，神经元应有的邻域关系，神经元的竞争与感受野的划分等问题，最后定量分析了任意的外部输入概率分布下，一维神经元阵列稳定权重的概率密度分布的递推关系式，最终还得到了外部输入服从均匀分布下的稳定权重的显式解。
The training process of Kohonen neural network is firstly divided into two phases,the formation phase of self-organizing topology preserve mapping and the post-training phase.Then an investigation is made into the topology preserve property,the neighborhood relationship,the competition and the division of receptive field of the neurons in the post-training phase.Furthermore,the quantitative analysis of probability distribution is given to the one dimensional neural array in the case of arbitrary distribution input,and the iterative formulas of the stated weights of neurons are obtained. Finally,the explicit solution of the static weights is presented to the uniform distributed input.
出处
《电子学报》
EI
CAS
CSCD
北大核心
1995年第8期52-56,共5页
Acta Electronica Sinica
基金
国家自然科学基金
