Fuzzy Adaptive Multitask Optimization.

Area of Science:

• Artificial Intelligence
• Computational Intelligence
• Optimization

Background:

• Evolutionary multitask optimization (EMTO) aims to optimize multiple tasks concurrently.
• Existing EMTO algorithms often use fixed knowledge transfer probabilities and single search operators, limiting adaptability.
• Fuzzy systems offer adaptability for complex, interdependent problems like EMTO.

Purpose of the Study:

• To propose a Fuzzy Adaptive Multitask Optimization (FAMTO) framework.
• To develop adaptive strategies for knowledge transfer and evolutionary search operators.
• To improve the performance and applicability of EMTO.

Main Methods:

• Implemented a Fuzzy Adaptive Transfer (FAT) strategy for adaptive knowledge transfer probability adjustment based on offspring survival and quality.
• Employed fuzzy logic to manage interdependent indicators for robust transfer probability tuning.
• Introduced an Individual-based Random Selection (IRS) strategy for adaptive intra-task evolutionary search operator selection.

Main Results:

• FAMTO demonstrated significantly superior performance compared to state-of-the-art EMTO algorithms on CEC17 and CEC22 benchmarks.
• The proposed method showed practical applicability when applied to a real-world planar kinematic arm control problem.
• Extended experiments on many-task optimization problems (MaTOPs) confirmed FAMTO's scalability.

Conclusions:

• FAMTO effectively addresses the limitations of fixed parameters in traditional EMTO.
• The adaptive strategies enhance optimization performance and robustness across diverse tasks.
• FAMTO presents a scalable and applicable solution for complex multitask optimization challenges.