Least absolute deviations learning of multiple tasks

Abstract

In this paper, we propose a new multitask feature selection model based on least absolute deviations. However, due to the inherent nonsmoothness of \begin{document}$l_1 $\end{document} norm, optimizing this model is challenging. To tackle this problem efficiently, we introduce an alternating iterative optimization algorithm. Moreover, under some mild conditions, its global convergence result could be established. Experimental results and comparison with the state-of-the-art algorithm SLEP show the efficiency and effectiveness of the proposed approach in solving multitask learning problems.