On the Boundedness of Subgradients in Distributed Optimization
Kui Zhu, Hao Zhu, Yutao Tang · 2020
Subgradient methods have drawn much attention in distributed optimization literature during the last few years. In most existing papers, the subgradients of all related objective functions are assumed to be uniformly bounded. Classical normalized step sizes often need certain global information and thus fail to remove this assumption in a distributed manner. The main goal of this paper is to provide some variants of subgradient methods to solve a generic distributed optimization problem without such a boundedness assumption by using some componentwise normalized step sizes. The efficacy of our algorithms is verified by a numerical example.