一种提高Kernel PCA特征提取性能的核优化算法

An optimizing kernel algorithm for improving the performance of Kernel PCA feature extraction

基于核的主分量分析(Kernel PCA)能够提取数据的非线性特征,但其性能受核参数的影响非常大.提出了一种新的基于特征空间中非高斯分布估计的核参数优化算法.该方法基于Kernel PCA中最优的参数应能导致特征空间中数据具有高斯分布的思想,通过对特征空间中数据的非高斯性结构进行分析,从反面估计其对高斯分布的逼近程度.采用该方法对各种数据进行实验都有很好的效果,表明了该方法的有效性.

Kernel PCA can effectively extract the nonlinear features of data set.However,the performance of Kernel PCA feature extraction is strongly influenced by the parameter of kernel.For this reason,a novel kernel parameter optimizing algorithm based on the nongaussian distribution estimation in feature space is presented.Based on the idea that the optimized kernel parameter can make the mapped data in feature space be Guassian distribution,the nongaussian structure of the mapped data is analyzed,and then the approximation degree of the mapped data's distribution to the Gaussian distribution in feature space is estimated.The experiments show this method is very effective to any type of data.