General Kernel Optimization Model Based on Kernel Fisher Criterion

In this paper a general kernel optimization model based on kernel Fisher criterion (GKOM) is presented. Via a data-dependent kernel function and maximizing the kernel Fisher criterion, the combination coefficients of different kernels can be learned adaptive to the input data. Finally positive empirical results on benchmark datasets are reported.

[1]  Si Wu,et al.  Improving support vector machine classifiers by modifying kernel functions , 1999, Neural Networks.

[2]  M. Omair Ahmad,et al.  Optimizing the kernel in the empirical feature space , 2005, IEEE Transactions on Neural Networks.