An Adaptive Stochastic Dominant Learning Swarm Optimizer for High-Dimensional Optimization.

Area of Science:

• Optimization algorithms
• Computational intelligence
• Swarm intelligence

Background:

• High-dimensional problems are prevalent across scientific and engineering fields.
• Existing optimization methods often struggle with efficiency and effectiveness in high dimensions.
• Balancing convergence speed and swarm diversity remains a key challenge.

Purpose of the Study:

• To propose a simple, efficient, and effective stochastic dominant learning swarm optimizer (SDLSO).
• To address the computational cost and complexity associated with high-dimensional optimization.
• To improve solution quality, convergence speed, and scalability for challenging problems.

Main Methods:

• Introducing a novel update mechanism where particles update only when dominated by randomly selected exemplars.
• Implementing a strategy to maintain swarm diversity by allowing implicit direct entry into the next generation.
• Developing an adaptive parameter adjustment strategy based on individual particle evolutionary information.
• Conducting extensive experiments on high-dimensional benchmark datasets.

Main Results:

• The proposed SDLSO demonstrates competitive or superior performance compared to state-of-the-art methods.
• Achieved excellent results in solution quality, convergence speed, scalability, and computational cost.
• Showcased particular effectiveness on partially separable and multimodal problems.
• Validated effectiveness through application to real-world feature selection problems.

Conclusions:

• The SDLSO offers a promising approach for tackling high-dimensional optimization challenges.
• The optimizer effectively balances exploration (diversity) and exploitation (convergence).
• Its adaptive nature and efficient update mechanism contribute to its strong performance on complex and real-world problems.