An active set Barzilar-Borwein algorithm for \(l_0\) regularized optimization (Q2307750)

The paper investigates a class of \(l_0\)-regularized optimization problems. After defining 4 kinds of stationary points (basic stationary points, strong stationary ponts, L-stationary points, strict L-stationay points) and their properties, it proposes an active set method able to approximate the correct support of a strict L-stationary point, also providing guarantees on its approximation accuracy (Theorems 3.1-3.3). Then, an algorithm is proposed for which the following property holds: any limit point of the sequence generated by such an algorithm converges to a strong stationary point (Theorems 4.1--4.3). Numerical results show the effectiveness of the algorithm proposed in the article, when it is embedded in a continuation procedure.