EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

3447. Avoiding strict saddle points of nonconvex regularized problems

Invited abstract in session TD-41: Lower-order composite optimization problems, stream Nonsmooth Optimization.

Tuesday, 14:30-16:00
Room: 97 (building: 306)

Authors (first author is the speaker)

1. Hao Wang
Information Science and Technology, ShanghaiTech University

Abstract

In this paper, we consider a class of non-convex and non-smooth sparse optimization problems, which encompass most existing nonconvex sparsity-inducing terms. These problems, moreover, are probably non-lipschitz around sparse solutions. We propose an damped iterative reweighted l1 algorithm to solve these problems. The algorithm is guaranteed to converge only to local minimizers when randomly initialized. By deriving a second-order sufficient optimal condition under the non-degenerate assumption, we show that the strict saddle property is generic on these sparse optimization problems.

Keywords

Status: accepted


Back to the list of papers