Parallel Majorization Minimization with Dynamically Restricted Domains for Nonconvex Optimization

We propose an optimization framework for nonconvex problems based on majorizationminimization that is particularity well-suited for parallel computing. It reduces the optimization of a high dimensional nonconvex objective function to successive optimizations of locally tight and convex upper bounds which are additively separable into low dimensional objectives. The original problem is then broken into simpler parallel sub-problems while still guaranteeing the monotonic reduction of the original objective function and convergence to a local minimum. Due to the low dimensionality of each sub-problem, secondorder optimization methods become computationally feasible and can be used to accelerate convergence. In addition, the upper bound can be restricted to a local dynamic convex domain, so that it is better matched to the local curvature of the objective function, resulting in accelerated convergence.