基于核函数的最大间隔聚类算法

Kernel-based maximal-margin clustring algorithm

提出了基于最优超平面与支持向量机思想的最大间隔聚类算法。该方法借鉴了最优超平面思想和用核函数非线性映射构造支持向量机的思想。通过构造一个二次规划问题 ,得到了使分类后两类间距最大的聚类方法 ,并且借助非线性核函数将该方法推广到非线性情况。仿真试验表明 :该方法可以较好地解决很多非监督分类问题 。

The paper proposed a new method of clustering with maximal margin based on the idea of optimal hyperplane and support vector machine. Through adopting the ideas of optimal hyperplan and nonlinear mapping of SVM, we construct the linear clustering method which makes the distance between two separated groups maximal by solving a quadratic programming problem. This linear method was generalized to nonlinear case using kernel functions. The result of experiment shows that it can deal with the unsupervised learning problem effectively.