Using Kernel Fisher Criterion for Gaussian Kernel Optimization
Li Zhong, Yu Ze Liu · Applied Mechanics and Materials · 2015
Empirical success of kernel-based learning methods is very much dependent on the kernel used. We propose an effective Gaussian kernel optimization approach for support vector machine (SVM). The key property of the proposed approach is that it adopts the kernel Fisher criterion (KFC) as the evaluation criterion to measure the goodness of the kernel used. After introducing a distance-based representation of KFC, we optimize the Gaussian kernel by using a gradient-based algorithm, which is based on the possibility of computing the gradient of KFC with respect to the width parameter of Gaussian kernel. The proposed approach is demonstrated with two popular UCI machine learning benchmark examples.