Multilayer Perceptron Grouping and Sparse Gaussian Process-Based Surrogate-Assisted Evolutionary Algorithm for...

Area of Science:

• Computational intelligence
• Optimization algorithms
• Machine learning

Background:

• Gaussian processes (GPs) are valuable for expensive optimization problems (EOPs) due to uncertainty quantification.
• The cubic computational complexity of GPs limits their scalability with increasing data.
• High-dimensional expensive multiobjective optimization problems (EMOPs) pose significant computational challenges.

Purpose of the Study:

• To develop a computationally efficient surrogate-assisted evolutionary algorithm (SAEA) for EMOPs.
• To overcome the scalability limitations of traditional Gaussian processes in high-dimensional spaces.
• To improve the balance between exploration and exploitation in optimization.

Main Methods:

• Integration of multilayer perceptron (MLP) grouping for subspace selection.
• Application of sparse Gaussian process (GP) models with optimized pseudo-input points for each objective function.
• Development of an adaptive sparse and diverse (ASD) infill criterion based on sparse GP predictive distributions.

Main Results:

• The proposed MLPSGP-SAEA demonstrates significant competitive advantages over existing state-of-the-art SAEAs.
• Experimental results on benchmark suites and an aerodynamic design problem validate the algorithm's effectiveness.
• The MLP grouping effectively reduces dimensionality, and sparse GPs enhance computational efficiency and accuracy.

Conclusions:

• MLPSGP-SAEA offers a computationally efficient and accurate solution for high-dimensional EMOPs.
• The integration of MLP grouping and sparse GPs effectively addresses the limitations of traditional GPs.
• The ASD infill criterion aids in balancing exploration and exploitation for improved optimization performance.