Title: Inertial block majorization minimization for solving multiblock
composite optimization problems

DuyNhat Phan

Abstract: In this talk, we present inertial block Majorization Minimization
(MM), a method for solving multiblock composite optimization problems with
or without coupling linear constraints. The idea of MM is to iteratively
minimize a surrogate function that locally approximates the objective
function of the optimization problem. By using suitable surrogate functions,
MM can recover well-known first-order algorithms such as the proximal point
and proximal gradient. In our study, we investigate sub-sequential
convergence as well as global convergence for the generated sequence of the
method. We also demonstrate the acceleration effects of the inertial
technique on two important machine learning problems, namely a matrix
completion problem, and a nonconvex low-rank representation problem.