The multiproximal linearization method for convex composite problems

TL;DR

This paper introduces the Multiprox method, an algorithm for convex composite minimization that linearizes smooth components and solves simple subproblems, providing explicit complexity bounds and handling moving constraints.

Contribution

The paper proposes the Multiprox method, a novel approach that generalizes existing algorithms and offers explicit complexity results for convex composite problems with moving constraints.

Findings

Explicit $O(1/k)$ complexity bounds derived.

Method encompasses moving balls, proximal Gauss-Newton, and forward-backward splitting.

Numerical experiments demonstrate effectiveness on complex geometries.

Abstract

Composite minimization involves a collection of smooth functions which are aggregated in a nonsmooth manner. In the convex setting, we design an algorithm by linearizing each smooth component in accordance with its main curvature. The resulting method, called the Multiprox method, consists in solving successively simple problems (e.g. constrained quadratic problems) which can also feature some proximal operators. To study the complexity and the convergence of this method we are led to study quantitative qualification conditions to understand the impact of multipliers on the complexity bounds. We obtain explicit complexity results of the form $O(\frac{1}{k})$ involving new types of constant terms. A distinctive feature of our approach is to be able to cope with oracles involving moving constraints. Our method is flexible enough to include the moving balls method, the proximal Gauss-Newton's method, or the forward-backward splitting, for which we recover known complexity results or establish new ones. We show through several numerical experiments how the use of multiple proximal terms can be decisive for problems with complex geometries.