Recursive Kernel MPE Loss Algorithm

Abstract

Kernel mean p-power error (KMPE) as a robust learning loss has been successfully employed to design robust PCA and ELM. This paper proposes a novel recursive adaptive filtering algorithm via the KMPE loss to identify linear system parameters under non-Gaussian noise cases. To derive the recursive KMPE algorithm, a KMPE loss with a forgetting factor is given first, and then the gradient method is employed to derive a recursive form of the weight estimation with a gain matrix for the system. Numerical simulation results demonstrate that the proposed algorithm with a suitable p value can obtain higher steady-state accuracy and faster convergence rate compared with some other existing algorithms.