Improving Estimation of Distribution Algorithm on Multimodal Problems by Detecting Promising Areas.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Machine Learning

Background:

• Estimation of Distribution Algorithms (EDAs) struggle with multimodal problems.
• Existing EDAs often require extensive function evaluations for exploration.
• A need exists for methods to enhance EDA efficiency on complex optimization landscapes.

Purpose of the Study:

• To introduce a novel multiple sub-models maintenance technique, MAPS.
• To improve the performance of EDAs on multimodal optimization problems.
• To accelerate the optimization speed of EDAs by reducing function evaluations.

Main Methods:

• Developed the Maintaining and Processing Sub-models (MAPS) technique.
• Integrated MAPS with three different types of EDAs that use a single Gaussian model.
• Assessed MAPS performance through empirical studies on 12 benchmark problems.

Main Results:

• MAPS explicitly detects promising areas, saving function evaluations.
• MAPS significantly accelerates convergence speed compared to standard EDAs.
• MAPS yields more stable solutions across various benchmark problems.

Conclusions:

• MAPS is an effective technique for enhancing EDAs on multimodal problems.
• The explicit detection of promising areas is key to MAPS's efficiency.
• MAPS offers a generalizable approach for improving EDA performance.