摘要
依据RBF神经元模型的几何解释,提出一种新的构造型神经网络分类算法.首先从样本数据本身入手,通过引入一个密度估计函数来对样本数据进行聚类分析;然后在特征空间里构造超球面,以逼近样本点分布的几何轮廓,从而将神经网络训练问题转化为点集“包含”问题.该算法有效克服了传统神经网络训练时间长、学习复杂的缺陷,同时也考虑了神经网络规模的优化问题.实验证明了该算法的有效性.
According to the geometrical representation of RBF neural model, a classification algorithm is proposed. Starting with the sample data directly, clustering analysis is proceeded by introducing a density function. And then hyperspheres are constructed to draw up the distribution of the sample data in feature space. The training problem of neural networks can be transformed into tbe “including” problem of a point set. The proposed algorithm can reduce the long training time and learning complexity of traditional neural networks. At the same time, the optimization of the neural network is also considered and computer simulation results show that the proposed neural network is quite efficient.
出处
《控制与决策》
EI
CSCD
北大核心
2005年第12期1411-1414,共4页
Control and Decision
