Multitask Diffusion Least-Mean-Fourth Algorithm

Abstract

In some applications, the multitask network may be corrupted by non-Gaussian noise, e.g., uniform noise or binary noise. If the multitask diffusion LMS algorithm is used in such situations, its steady-state performance will be degraded. To overcome this issue, this work presents a multitask diffusion version of the least-mean-fourth algorithm by using the fourth-order moment of the estimation error. To further enhance its convergence rate, the $l_{0}$-norm regularization is used. Simulation results show that our algorithms can obtain small steady-state mean-square deviation (MSD).