Manifold Interpolation for Large-Scale Multiobjective Optimization via Generative Adversarial Networks.

Area of Science:

• Computational intelligence
• Optimization algorithms
• Machine learning

Background:

• Large-scale multiobjective optimization problems (LSMOPs) present challenges due to numerous variables and conflicting objectives.
• Existing algorithms struggle with Pareto optimal solution (POS) manifold learning, leading to poor diversity and inefficient searches.
• Traditional genetic operators exhibit deficiencies in handling the complex distribution of POSs in LSMOPs.

Purpose of the Study:

• To propose a novel generative adversarial network (GAN)-based manifold interpolation framework for LSMOPs.
• To enhance the learning of the POS manifold and generate high-quality solutions.
• To improve the performance of evolutionary algorithms in solving LSMOPs.

Main Methods:

• Development of a GAN-based framework for manifold interpolation.
• Learning the low-dimensional manifold of Pareto optimal solutions.
• Integration of the GAN framework with evolutionary algorithms to generate superior solutions.

Main Results:

• The proposed GAN-based framework demonstrates significant improvements in solving LSMOPs.
• Experimental results show enhanced performance compared to state-of-the-art algorithms on benchmark functions.
• The approach effectively addresses issues of poor diversity and local optima.

Conclusions:

• The GAN-based manifold interpolation framework offers a promising approach for tackling LSMOPs.
• This method enhances the ability of evolutionary algorithms to find high-quality Pareto optimal solutions.
• The framework contributes to advancing the field of large-scale multiobjective optimization.