CAJ | 학술논문

根据Kohonen自组织特征映射神经网络中学习阶段的性质,对标准Kohonen神经网络进行了改进,通过选取不同的学习率和邻域函数,将学习阶段分为粗调整学习和微调整学习二个阶段,使Kohonen神经网络的训练速度和收敛效果有一定程度的提高,改善了聚类效果.同时在Kohonen神经网络的输出层添加一模糊聚类层,为挖掘用户的多种兴趣提供了一种可行的方法.
근거Kohonen자조직특정영사신경망락중학습계단적성질,대표준Kohonen신경망락진행료개진,통과선취불동적학습솔화린역함수,장학습계단분위조조정학습화미조정학습이개계단,사Kohonen신경망락적훈련속도화수렴효과유일정정도적제고,개선료취류효과.동시재Kohonen신경망락적수출층첨가일모호취류층,위알굴용호적다충흥취제공료일충가행적방법.
According to the Kohonen self-organizing feature map neural network in the nature of the learning stage,the studying will be divided into two stages by choosing a different learning rate and neighborhood function: rough adjustment studying and micro-adjustment studying.It leads to improve the results from Kohonen Neural network convergence and the clustering accuracy,and provides a workable solution for a variety of users interesting in customers mining.