In this paper, we study the iteration complexity of a blockcoordinategradientdescent (BCGD) method with a cyclic rule for solving convex optimization problems. We propose a new Lipschitz continuity-like assumption ...
详细信息
In this paper, we study the iteration complexity of a blockcoordinategradientdescent (BCGD) method with a cyclic rule for solving convex optimization problems. We propose a new Lipschitz continuity-like assumption and show that the iteration complexity for the proposed BCGD method can be improved to O(max{M, L}/epsilon), where M is the constant in the proposed assumption, L is the usual Lipschitz constant for the gradient of the objective function, and epsilon > 0 is the required precision. In addition, we analyze the relation between M and L, and prove that, in the worst case, M <= root NL, where N is the number of blocks.
暂无评论