| Okkes Tolga Altinoz Redesigning the Particle Swarm Optimization Algorithm as a Reinforcement Learning Methodology |
|---|
| Abstract. Modern optimization algorithms consist of algorithms that have been developed, improved, and expanded for different problems with specific properties as extensive, stochastic, noise multi, many, bi-level, as well as traditional optimization problems. These relatively complex modern algorithms are mainly based on traditional numerical methods. Their complex structures are increasingly being extended with learning-based methods as they can be used in conjunction with machine learning methods, particularly reinforcement learning, and are also used within these methods. These algorithms can be evolutionary or nature-inspired algorithms. Among nature-inspired algorithms, one of the most widely used is the Particle Swarm Optimization (PSO) algorithm. The PSO algorithm has been used in conjunction with machine learning algorithms for solving the problem-oriented applications; and Reinforcement Learning as a machine learning method contains potential to use with PSO. Therefore, in this study, the PSO algorithm will be redesigned as a reinforcement learning framework. The PSO in a different framework will be evaluated on benchmark problems where the innovation suggestions will be made regarding the parameters and formulations of the algorithm according to their areas of use, its performance will be evaluated, and improvement suggestions will be presented. |
| Keywords: Particle Swarm Optimization, Reinforcement Learning, Learning-based Optimization, Machine Learning, Single-objective, Optimization |
 Download PDF |
| DOI: https://doi.org/10.54381/itta2026.1.02 |