Abstract
In the context of Quadratically Constrained Quadratic Programming (QCQP) with dynamic parameters, the effectiveness of various optimization approaches is heavily influenced by the quality of the initial guess. To address this challenge, this paper proposes a novel approach that leverages reinforcement learning (RL) to generate high-performing initial guesses for iterative algorithms, with the dynamic parameters serving as inputs. Our approach aims to accelerate convergence and improve the objective value, thereby enabling efficient and effective solutions to the QCQP problem under variability. In this study, we evaluate the proposed approach by applying it to an iterative algorithm, specifically the Iterative Rank Minimization (IRM) algorithm. Our empirical evaluations demonstrate the efficacy of the proposed approach in solving QCQP problems with dynamic parameters. The RL-guided IRM algorithm yields high-quality solutions, as evidenced by significantly improved optimality and faster convergence when compared to the original IRM algorithm.
Recommended Citation
C. Pei et al., "Reinforcement Learning-Guided Quadratically Constrained Quadratic Programming for Enhanced Convergence and Optimality," Proceedings of the IEEE Conference on Decision and Control, pp. 7293 - 7298, Institute of Electrical and Electronics Engineers, Jan 2023.
The definitive version is available at https://doi.org/10.1109/CDC49753.2023.10383301
Department(s)
Mechanical and Aerospace Engineering
International Standard Serial Number (ISSN)
2576-2370; 0743-1546
Document Type
Article - Conference proceedings
Document Version
Citation
File Type
text
Language(s)
English
Rights
© 2024 Institute of Electrical and Electronics Engineers, All rights reserved.
Publication Date
01 Jan 2023
Comments
National Science Foundation, Grant CPS-2201568