多核模糊聚类  

Multiple kernel fuzzy clustering

作　　者：戴思薇[1] 吴小俊[1] 高翠芳[1] 

机构地区：[1]江南大学物联网工程学院,江苏无锡214122

出　　处：《计算机工程与应用》2016年第2期65-69,111,共6页Computer Engineering and Applications

基　　金：111引智计划基金(No.B12018)

摘　　要：针对单核聚类的性能局限性问题,提出将高斯核、Sigmoid核以及多项式核等多种核组成一种新的多核函数,并利用于模糊核进行聚类。高斯核在聚类中有广泛应用,同时Sigmoid核在神经网络中被证明具有很好的全局分类性能。将不同的核函数组合起来的多核函数将结合各种核函数的优点,其聚类性能优于利用单核的模糊核聚类(KFCM),实验结果表明了该方法的有效性。Because the fuzzy clustering based on single kernel function has some limitation on performance, a new multiplekernel fuzzy clustering is put forward. The new multiple kernel is constituted by Gaussian kernel, Sigmoid kernel and polynomial kernel. Gaussian kernel is wildly used in KFCM. It can be demonstrated that sigmoid kernel derived from neural network has good global classification performance, as well as polynomial kernel. The new multiple kernel combines the advantage of them. The experimental results prove that the fuzzy clustering with multiple kernel is much better than with single kernel on performance.

关 键 词：多核 模糊核聚类 高斯核 Sigmoid核 多项式核函数 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]