摘要
在人工神经网络中,拓扑函数用于生成神经元的空间拓扑结构。不同的拓扑函数可能导致神经网络形成不同的学习结果。以Matlab的拓扑函数模板为基础,建立了四种拓扑函数:cossintopf、sincostopf、acossintopf及expsintopf。同时,以这些拓扑函数生成神经元拓扑结构,用于自组织特征映射网络对田间无脊椎动物目的无监督自组织聚类。结果显示,聚类各有差异,可根据与实际情形的比较来选择拓扑函数与聚类结果,或进行综合分析。
Topological functions are always used to generate the spatial structure of neurons in artificial neural networks （ANNs）. Different topological functions will probably yield various results in the learning of ANNs. Four topological functions, i. e. , cossintopf, sincostopf, acossintopf, and expsintopf,were developed based on template in toolbox of Matlab. We used these functions in Self-Organizing Feature Map and conducted the non-supervisory clustering of invertebrates orders in rice field. Results showed that clusters are different when using different topological functions. We may chose these functions and results based on comparison with the practical situation.
出处
《计算机应用与软件》
CSCD
北大核心
2006年第10期71-73,共3页
Computer Applications and Software
基金
国家自然科学基金(NO.30170184)
教育部留学回国人员科研基金(2000)
亚洲开发银行项目(RETA5711)
