An information theoretic kernel algorithm for robust online learning

Abstract

Kernel methods are widely used in nonlinear modeling applications. In this paper, a robust information theoretic sparse kernel algorithm is proposed for online learning. In order to reduce the computational cost and make the algorithm suitable for online applications, we investigate an information theoretic sparsification rule based on the mutual information between the system input and output to determine the update of the dictionary (support vectors). According to the rule, only novel and informative samples are selected to form a sparse and compact dictionary. Furthermore, to improve the generalization ability, a robust learning scheme is proposed to avoid the algorithm over learning the redundant samples, which assures the convergence of the learning algorithm and makes the learning algorithm converge to its steady state much faster. Experiment are conducted on practical and simulated data and results are shown to validate the effectiveness of our proposed algorithm.