An active learning Gaussian modeling based multi-objective evolutionary algorithm using population guided weight...

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Machine Learning

Background:

• Multi-objective evolutionary algorithms (MOEAs) often rely on crossover operators for offspring generation.
• Existing MOEAs face challenges in adapting to diverse Pareto front (PF) shapes.
• Inverse model-based MOEAs (IM-MOEA) offer an alternative computing schema.

Purpose of the Study:

• To propose an active learning Gaussian modeling based multi-objective evolutionary algorithm (ALGM-MOEA).
• To enhance search efficiency and prediction accuracy in MOEAs.
• To effectively handle multi-objective problems with varied PF shapes.

Main Methods:

• Developed a population-guided weight vector evolution strategy for dynamic adjustment of search directions based on PF distribution.
• Implemented an active learning-based training sample selection for Gaussian process inverse models.
• Utilized Gaussian process models to predict offspring generation.

Main Results:

• The proposed population-guided weight vector evolution strategy effectively adapts to different PF shapes.
• Active learning enhances the prediction accuracy of Gaussian process inverse models.
• ALGM-MOEA demonstrates competitive performance on benchmark multi-objective optimization problems.

Conclusions:

• ALGM-MOEA offers a robust approach for multi-objective optimization across various PF shapes.
• The integration of active learning and population-guided strategies improves MOEA performance.
• The proposed method provides a competitive alternative to traditional MOEA designs.