AIEA: An Asynchronous Influence-Based Evolutionary Algorithm for Expensive Many-Objective Optimization.

Area of Science:

• Computational Science
• Optimization Theory
• Machine Learning

Background:

• Expensive multi/many-objective optimization problems (EMOPs) involve computationally intensive objective evaluations, often from diverse simulation tools with varying latencies.
• Serial optimization of EMOPs leads to prohibitive computational costs.
• Parallel surrogate modeling offers a promising avenue for efficiency, but challenges remain in model accuracy and effective candidate selection.

Purpose of the Study:

• To propose an efficient asynchronous surrogate-assisted evolutionary algorithm (SAEA) for EMOPs.
• To enhance model accuracy and candidate selection in parallel surrogate modeling for EMOPs.
• To address the challenges of computational cost and scalability in solving EMOPs.

Main Methods:

• A client-server architecture is employed, where clients handle objective approximation and the server manages evolution.
• An 'influence degree' is introduced for adaptive candidate selection in the objective space.
• A 'most-uncertain-first' strategy guides asynchronous evaluations and model improvement.
• Nearest neighbor inheritance is used to handle incomplete objective values.

Main Results:

• The proposed asynchronous influence-based SAEA (AIEA) demonstrates improved global optimization capabilities.
• Experimental comparisons show AIEA outperforms five other surrogate-assisted evolutionary algorithms.
• The algorithm exhibits strong scalability for complex EMOPs.

Conclusions:

• AIEA effectively addresses the computational challenges of EMOPs through parallel surrogate modeling and intelligent candidate selection.
• The influence-based approach and asynchronous strategy significantly enhance optimization efficiency and accuracy.
• AIEA offers a scalable and robust solution for expensive multi-objective optimization tasks.