Affine Transformation-Enhanced Multifactorial Optimization for Heterogeneous Problems.

Area of Science:

• Evolutionary Computation
• Optimization Algorithms
• Machine Learning

Background:

• Evolutionary multitasking (EMT) enhances convergence by transferring knowledge across optimization tasks.
• Current EMT algorithms struggle with heterogeneous problems lacking similar fitness landscapes.
• Existing domain adaptation methods for EMT often suffer from intertask mapping degradation.

Purpose of the Study:

• To propose a novel rank loss function for superior intertask mapping in EMT.
• To generalize EMT for solving heterogeneous optimization problems.
• To address the degradation of intertask mapping in cross-task knowledge transfer.

Main Methods:

• Developed a novel rank loss function to improve intertask mapping.
• Utilized an evolutionary-path-based representation model.
• Derived an analytical solution for affine transformation to bridge distinct problem domains.
• Integrated a mapping-based transferability enhancement technique into the EMT paradigm.

Main Results:

• The proposed rank loss function facilitates superior intertask mapping.
• The method effectively bridges the gap between heterogeneous optimization tasks.
• Experimental validation demonstrated the efficacy against state-of-the-art EMT algorithms.
• Successful application on synthetic benchmarks and a practical case study.

Conclusions:

• The novel rank loss function and mapping-based technique significantly enhance EMT for heterogeneous problems.
• This approach overcomes limitations of existing methods by preventing intertask mapping degradation.
• The proposed technique offers a robust and generalizable solution for cross-task knowledge transfer in evolutionary computation.