A modified estimation distribution algorithm based on extreme elitism.

Area of Science:

• Artificial Intelligence
• Optimization Algorithms
• Machine Learning

Background:

• Existing estimation distribution algorithms (EDAs) with univariate marginal Gaussian models face challenges in balancing convergence speed and population diversity.
• Premature convergence can limit the effectiveness of standard EDAs in complex optimization tasks.

Purpose of the Study:

• To improve the performance of estimation distribution algorithms (EDAs) by introducing a novel extreme elitism selection method.
• To enhance the convergence rate and maintain population diversity in EDAs for better optimization outcomes.

Main Methods:

• Developed and integrated an extreme elitism selection strategy into an existing estimation distribution algorithm (EDA).
• The modified EDA was evaluated using benchmark low-dimensional and high-dimensional optimization problems.
• Analyzed the impact of the new selection method on EDAs, considering the no-free-lunch theorem.

Main Results:

• The extreme elitism selection method significantly accelerated the convergence rate of the EDA.
• The proposed method effectively preserved population diversity, preventing premature convergence.
• Empirical results demonstrated the superior performance of the modified EDA on various optimization problems.

Conclusions:

• The extreme elitism selection is a valuable enhancement for estimation distribution algorithms (EDAs).
• This approach offers a robust strategy for tackling complex optimization challenges by balancing exploration and exploitation.
• The findings provide insights into improving evolutionary computation techniques.