A Random Particle Swarm Optimization Based on Cosine Similarity for Global Optimization and Classification Problems.

Area of Science:

• Optimization Algorithms
• Computational Intelligence
• Machine Learning

Background:

• The increasing complexity of optimization problems necessitates algorithms with superior global optimization capabilities.
• Traditional Particle Swarm Optimization (PSO) faces challenges in effectively exploring vast search spaces and achieving precise local optima.
• Developing advanced optimization techniques is crucial for various scientific and engineering applications.

Purpose of the Study:

• To introduce a novel optimization algorithm, Random Particle Swarm Optimization (RPSO), designed for enhanced global optimization.
• To improve the exploration and exploitation balance within the PSO framework.
• To validate the performance of RPSO on benchmark datasets and real-world applications like Convolutional Neural Network (CNN) classification.

Main Methods:

• Developed RPSO by enhancing the traditional PSO with adapted parameter selection and a Random Contrastive Interaction (RCI) mechanism.
• Incorporated quadratic interpolation (QI) to improve local search efficiency.
• Utilized cosine similarity for dynamic population information updates in both RCI and QI selection.

Main Results:

• RPSO demonstrated strong competitiveness against state-of-the-art algorithms on the IEEE Congress on Evolutionary Computation (CEC) 2022 test dataset.
• The algorithm showed significant improvements in global optimization task performance.
• In CNN-based medical image classification, RPSO achieved comparable stability and accuracy to existing methods.

Conclusions:

• RPSO offers an effective approach to tackling complex global optimization problems.
• The enhancements in RPSO, including RCI and QI, lead to more efficient search and improved solution quality.
• RPSO shows promise for enhancing the performance of machine learning models, particularly in CNN classification tasks.