An online learning algorithm of support vector regression based on natural gradient

Support vector regression based on the quadratic programming is unfit for the online training and predicting, and this paper proposes an online algorithm of the support vector regression based on the natural gradient. The algorithm resolves the slow convergence of the standard gradient descent method by the plateau phenomenon, and increases learning speed. And its dynamical behavior is proved to be Fisher efficient, implying that it has the same performance as the optimal batch estimation of parameters. The results of experiments show it is an efficient online algorithm of the support vector regression.