EUROPT 2024
Abstract Submission

3. A feasible directions method for nonconvex optimization over linear constraints with a nonsmooth concave regularizer

Invited abstract in session TD-5: Nonsmooth optimization algorithms, stream Nonsmooth and nonconvex optimization algorithms.

Thursday, 14:10 - 15:50
Room: M:N

Authors (first author is the speaker)

1. Nadav Hallak
The Technion
2. Amir Beck
School of Mathematical Sciences, Tel-Aviv University

Abstract

This talk presents a feasible directions approach for the minimization of a continuous function over linear constraints in which the update directions belong to a predetermined finite set spanning the feasible set.
These directions are recurrently investigated in a cyclic semi-random order, where the stepsize of the update is determined via univariate optimization. The method achieves that any accumulation point is a stationary point, and enjoys a sublinear rate of convergence in expectation w.r.t a new optimality measure that acts as a proxy for the stationarity condition.

Keywords

Status: accepted


Back to the list of papers