Mirror descent and convex optimization problems with non-smooth inequality constraints (Q2415205)

Problems consisting in minimizing convex functions subject to convex nonsmooth inequality constraints appear in various applications and there are only a few computational methods that can solve them, usually in particular cases and frameworks. In this paper, the authors propose mirror-descent-type methods for approaching such problems in different frameworks and under various hypotheses such as Lipschitz continuity/(non)smoothness of the objective function, (strong) convexity of the involved functions, and deterministic/randomized information about the latter. In particular, a mirror descent scheme with adaptive stepsizes and an adaptive stopping rule is proposed. The question of recovering the dual solution in the considered problem is also discussed as well as the extension of the method to a composite framework. For the entire collection see [Zbl 1407.90006].