472. A full splitting algorithm for structured difference-of-convex programs
Invited abstract in session WC-6: Structured nonsmooth optimization -- Part III, stream Nonsmooth and nonconvex optimization.
Wednesday, 14:00-16:00Room: B100/7013
Authors (first author is the speaker)
| 1. | Rossen Nenov
|
| Acoustic Research Institute, Austrian Academy of Sciences |
Abstract
We address a class of nonconvex and nonsmooth problems in which the objective function is formed as the sum of a smooth function and the difference of two convex function composed with different linear operators. This structure gives rise to challenges from both nonsmoothness and nonconvexity. To tackle this, we introduce an adaptive double-proximal, full-splitting algorithm that separates the linear mappings from the nonsmooth terms through a moving center technique in the final subproblem. Our analysis establishes that the iterates converge subsequentially to an approximate stationary point and, under the Kurdyka-Łojasiewicz property, achieve global convergence. We also provide a counterexample demonstrating that the search for exact solutions can lead to divergence, underscoring the necessity of an approximate approach.
This is joint work with Radu Bot and Min Tao.
Keywords
- Non-smooth optimization
Status: accepted
Back to the list of papers