A block inertial Bregman proximal algorithm for nonsmooth nonconvex problems
share
In this paper, a block inertial Bregman proximal algorithm, namely [], for solving structured nonsmooth nonconvex optimization problems is introduced. More specifically, the objective function is the sum of a block relatively smooth function (i.e., relatively smooth with respect to each block) and block separable (nonsmooth) nonconvex functions. It is shown that the sequence generated by [] converges subsequentially to critical points, while it is globally convergent for Kurdyka-Łojasiewicz (KŁ) functions. Furthermore, the rate of convergence of this sequence is studied for the Łojasiewicz-type KŁ functions.
READ FULL TEXT