A New Approach To Online Regression Based On Maximum Correntropy Criterion

The problem of linear adaptive filtering (or equivalently, online regression) in the presence of non-Gaussian noise is addressed. One efficient way in face of environments with non-Gaussian noise is to employ information theoretic criteria such as correntropy. In this study, a new algorithm based on correntropy is proposed, and demonstrated to outperform previous works in terms of both convergence speed and steady-state misalignment. At the same time, the proposed algorithm benefits from lower computational complexity compared to some of these algorithms.