Retour

G-2021-12

A proximal quasi-Newton trust-region method for nonsmooth regularized optimization

, et

référence BibTeX

We develop a trust-region method for minimizing the sum of a smooth term \(f\) and a nonsmooth term \(h\), both of which can be nonconvex. Each iteration of our method minimizes a possibly nonconvex model of \(f + h\) in a trust region. The model coincides with \(f + h\) in value and subdifferential at the center. We establish global convergence to a first-order stationary point when \(f\) satisfies a smoothness condition that holds, in particular, when it has Lipschitz-continuous gradient, and \(h\) is proper and lower semi-continuous. The model of \(h\) is required to be proper, lower-semi-continuous and prox-bounded. Under these weak assumptions, we establish a worst-case \(O(1/\epsilon^2)\) iteration complexity bound that matches the best known complexity bound of standard trust-region methods for smooth optimization. We detail a special instance, named TR-PG, in which we use a limited-memory quasi-Newton model of \(f\) and compute a step with the proximal gradient method, resulting in a practical proximal quasi-Newton method. We establish similar convergence properties and complexity bound for a quadratic regularization variant, named R2, and provide an interpretation as a proximal gradient method with adaptive step size for nonconvex problems. R2 may also be used to compute steps inside the trust-region method, resulting in an implementation named TR-R2. We describe our Julia implementations and report numerical results on inverse problems from sparse optimization and signal processing. Both TR-PG and TR-R2 exhibit promising performance and compare favorably with two linesearch proximal quasi-Newton methods based on convex models.

, 30 pages

Ce cahier a été révisé en mars 2024

Axe de recherche

Application de recherche

Publication

, et
SIAM Journal on Optimization, 32(2), 900–929, 2022 référence BibTeX

Document

G2112R.pdf (750 Ko)

Matériel additionnel

G2112SMR.pdf (230 Ko)