EUROPT 2024
Abstract Submission

10. Learning to Relax Nonconvex Quadratically Constrained Quadratic Programs

Invited abstract in session WC-2: Conic and Semidefinite Optimization, stream Conic optimization: theory, algorithms and applications.

Wednesday, 10:05 - 11:20
Room: M:O

Authors (first author is the speaker)

1. Burak Kocuk
Industrial Engineering, Sabanci University
2. Buket Özen
Industrial Engineering, Sabancı University

Abstract

Nonconvex quadratically constrained quadratic programs are NP-Hard to solve in general. Literature primarily employs either semidefinite or linear programming to relax them, which are usually effective for distinct problem types, and a holistic understanding of which relaxation should be preferred over the other for a given instance is lacking. In this research, we present a learning-based approach to predict whether a semidefinite or linear programming relaxation would produce a stronger bound for a given instance by examining spectral properties and sparsity patterns of the data matrices.

Keywords

Status: accepted


Back to the list of papers