Sparse Nonnegative Learning for Multi-factor Model

This paper studies the identification of a multifactor model. To seek factors that play an important role among a large number of factors and ignore those negligible ones, the sparse multi-factor model is a reasonable choice. In addition, considering that non-negativity is a desired constraint imposed on the model parameters due to the inherent physical characteristics of the model in many real-world applications, the sparse nonnegative LMS (SNN-LMS) is applied to the identification of multi-factor model. Besides, the reweighted sparse nonnegative LMS (RSNN-LMS) is proposed to further improve the filtering performance. The reweighted scaling function makes its convergence rates more balanced, and the reweighted zero attractor helps shrinking coefficients selectively. Simulation results are presented to show the performance of the proposed algorithms.