CAJ | 학술논문

本文主要是研究自组织神经网络作为一种具有拓扑限制的，以特征提取为主要手段的聚类算法，并与传统的 K-means 算法进行比较分析，并将它们应用于几组人工数据。传统的 K-means 算法具有计算效率高的优点，但是聚类结果不稳定，初始值对于聚类收敛的结果有一定的影响，相比之下，自组织神经网络由于其引入具有拓扑结构的邻域函数，虽然计算效率比较低，但是可以达到较为稳定的聚类结果，且受初值影响较小。
본문주요시연구자조직신경망락작위일충구유탁복한제적，이특정제취위주요수단적취류산법，병여전통적 K-means 산법진행비교분석，병장타문응용우궤조인공수거。전통적 K-means 산법구유계산효솔고적우점，단시취류결과불은정，초시치대우취류수렴적결과유일정적영향，상비지하，자조직신경망락유우기인입구유탁복결구적린역함수，수연계산효솔비교저，단시가이체도교위은정적취류결과，차수초치영향교소。
This paper studies the clustering performance of self-organizing feature map,which is known as a topological reserving clustering algorithm in comparing with K-means on a set of artificial data. The conventional clustering method,K-means,is an efficient algorithm in dealing with quantitative data classification,but has been shown a weak and instable convergence to global optimal solution due to different initial settings. The self-organizing map is more robust in achieving the optimal solution due to its unique neighborhood function which has a topology constraint though it can slow down the learning speed. In addition,SOM learning is insensitive to initial values.