Adaptive Convex Combination of Kernel Maximum Correntropy Criterion

Over the past few years, the kernel maximum correntropy (KMC) algorithm has attracted much attention. But the traditional KMC uses a fixed correntropy-induced kernel width, which may result in undesirable performance if the kernel width is not appropriately selected. To overcome this shortcoming, we consider a convex combination of two KMC algorithms with different kernel widths. By adjusting the control parameter, the proposed algorithm can enjoy the advantages from two separate KMC algorithms. In addition, by applying some widely used assumptions, we conduct the convergence analysis, as well as the steady-state excessive mean-square error (EMSE) analysis. Simulations have demonstrated the superiority of our finding.