Multidomain Evolutionary Optimization on Combinatorial Problems in Complex Networks.

Area of Science:

• Artificial Intelligence
• Complex Systems
• Optimization

Background:

• Multitask evolutionary optimization (MTEO) focuses on task similarity for knowledge transfer.
• Real-world complex systems often share underlying characteristics (e.g., power-law, community structure).
• Exploiting these shared characteristics across different domains offers untapped potential for optimization.

Purpose of the Study:

• To introduce a novel framework, multidomain evolutionary optimization (MDEO), for knowledge transfer across diverse domains.
• To develop mechanisms for managing and effectively transferring solutions between different complex systems.
• To enhance evolutionary optimization by utilizing shared characteristics of complex systems.

Main Methods:

• Proposed a community-level measurement for graph similarity to guide knowledge transfer.
• Developed a graph-learning-based network alignment model for effective solution transfer.
• Devised a self-adaptive mechanism for determining the number of transferred solutions.
• Introduced a knowledge-guided mutation mechanism for adaptive utilization of transferred knowledge.

Main Results:

• The proposed multidomain evolutionary optimization (MDEO) framework demonstrates superior efficacy.
• Experiments conducted on adversarial link perturbation using real-world networks of different domains validated the framework's performance.
• MDEO outperformed classical evolutionary optimization methods.

Conclusions:

• Multidomain evolutionary optimization (MDEO) effectively leverages shared characteristics across different domains for enhanced optimization.
• The proposed framework provides a robust approach for knowledge transfer in complex systems.
• This research opens new avenues for cross-domain optimization in various scientific fields.