Individual-Based Transfer Learning for Dynamic Multiobjective Optimization.

Area of Science:

• Optimization
• Artificial Intelligence
• Computer Science

Background:

• Dynamic multiobjective optimization problems (DMOPs) involve time-varying functions, requiring rapid and accurate tracking of Pareto-optimal sets (POSs).
• Transfer learning is effective for DMOPs, but negative transfer can hinder optimization efficiency.
• Minimizing negative transfer is crucial for successful transfer learning applications in DMOPs.

Purpose of the Study:

• To propose a novel individual-based transfer learning method, the individual transfer-based dynamic multiobjective evolutionary algorithm (IT-DMOEA), for solving DMOPs.
• To enhance the efficiency and accuracy of solving DMOPs by mitigating negative transfer.

Main Methods:

• The IT-DMOEA employs a presearch strategy to filter high-quality, diverse individuals, preventing negative transfer from aggregated populations.
• An individual-based transfer learning technique is utilized to accelerate initial population construction.
• Combines transfer learning benefits with strategies to avoid negative transfer.

Main Results:

• The IT-DMOEA demonstrates improved solution quality and convergence speed compared to existing state-of-the-art algorithms.
• Experimental results validate the effectiveness of the proposed approach on benchmark DMOPs.

Conclusions:

• The IT-DMOEA effectively addresses the challenges of negative transfer in dynamic multiobjective optimization.
• The proposed method offers a significant advancement in solving DMOPs, improving both solution quality and computational efficiency.