论文信息 - A Study of Tuning Hyperparameters for Support Vector Machines

A Study of Tuning Hyperparameters for Support Vector Machines

Automatic parameters selection is an important issue to make support vector machines (SVMs) practically useful. Most existing approaches use Newton method directly to compute the optimal parameters. They treat parameters optimization as an unconstrained optimization problem. In this paper, the limitation of these existing approached is stated and a new methodology to optimize kernel parameters, based on the computation of the gradient of penalty function with respect to the RBF kernel parameters, is proposed. Simulation results reveal the feasibility of this new approach and demonstrate an improvement of generalization ability.

Jie Yang | Yong Quan | Chengzhou Ye

[1] Thorsten Joachims,et al. Estimating the Generalization Performance of an SVM Efficiently , 2000, ICML.

[2] V. Vapnik,et al. Bounds on Error Expectation for Support Vector Machines , 2000, Neural Computation.

[3] Sayan Mukherjee,et al. Choosing Multiple Parameters for Support Vector Machines , 2002, Machine Learning.

[4] K. Al-Sultan,et al. A new penalty function algorithm for convex quadratic programming , 1997 .

[5] Corinna Cortes,et al. Support-Vector Networks , 1995, Machine Learning.

[6] Bernhard Schölkopf,et al. Support vector learning , 1997 .

[7] Vladimir Vapnik,et al. Statistical learning theory , 1998 .

[8] Nello Cristianini,et al. An introduction to Support Vector Machines , 2000 .

[9] Nello Cristianini,et al. An Introduction to Support Vector Machines and Other Kernel-based Learning Methods , 2000 .

[10] Gunnar Rätsch,et al. An introduction to kernel-based learning algorithms , 2001, IEEE Trans. Neural Networks.