Learning-aided Artificial Bee Colony with neural knowledge transfer for global optimization.

Area of Science:

• Artificial Intelligence
• Swarm Intelligence
• Evolutionary Computation

Background:

• Swarm intelligence algorithms face challenges with effective knowledge utilization and negative knowledge transfer.
• Existing methods often struggle to prevent detrimental knowledge transfer across different search space regions.

Purpose of the Study:

• To introduce the Learning-Aided Artificial Bee Colony (LA-ABC) framework for global optimization.
• To develop a novel Neural Knowledge Transfer mechanism to enhance swarm intelligence.
• To mitigate negative knowledge transfer in optimization algorithms.

Main Methods:

• LA-ABC employs a dual-pathway architecture, combining swarm exploration with a knowledge-transfer pathway.
• An Artificial Neural Network (ANN) learns from historical successful solutions to model complex contexts.
• A generative operator transfers validated positive knowledge to create high-quality candidate solutions.

Main Results:

• LA-ABC demonstrated superior performance across 23 standard test functions and the IEEE CEC 2019 suite.
• The framework significantly outperformed 12 state-of-the-art algorithms, including ABC, L-SHADE, and RL variants.
• Rigorous statistical tests confirmed the significance of the improvements achieved by LA-ABC.

Conclusions:

• LA-ABC transforms swarm intelligence algorithms into intelligent agents capable of learning and navigating fitness landscapes.
• The proposed Neural Knowledge Transfer mechanism offers a robust paradigm for integrating reinforcement learning and knowledge transfer in evolutionary computation.
• LA-ABC provides a significant advancement in global optimization techniques.