Evolving Parameters of Multi-Scale Radial Basis Function Kernels for Support Vector Machines

T. Phienthrakul and B. Kijsirikul (Thailand)


Computational Intelligence, Evolutionary Strategies, Support Vector Machines, Radial Basis Function


In support vector machines (SVMs), kernel functions are used to compute dot product in a higher dimensional feature space. The performance of classification depends on the chosen kernel. The radial basis function (RBF) kernel is a most popular kernel that succeeded in many tasks. In order to obtain a more flexible kernel function, a family of RBF kernels is proposed. Multi-scale RBF kernels are combined by including weights. This new kernel is proved to be a Mercer's kernel. Then, the evolutionary strategies (ESs) are used to adjust the weights and the widths of the RBF kernels. The training accuracy, the bound of generalization error, and subsets cross-validation are used to be objective functions in the evolutionary process. The experimental results show that the proposed kernel allows better discrimination in the feature space. Moreover, subsets cross-validation is a good objective function and yields the effective results on benchmarks.

Important Links:

Go Back