Self-learning salp swarm algorithm for global optimization and its application in multi-layer perceptron model...

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Machine Learning

Background:

• Swarm intelligence algorithms are effective for optimization but struggle with complex, unknown landscapes due to fixed search strategies.
• Existing methods lack adaptability, limiting their application to diverse real-world optimization challenges.

Purpose of the Study:

• To introduce a novel self-learning mechanism to the Salp Swarm Algorithm (SSA), creating the Self-Learning Salp Swarm Algorithm (SLSSA).
• To enhance swarm intelligence for improved performance on varied and complex optimization problems.
• To develop an efficient black-box optimizer applicable to a wide range of problems.

Main Methods:

• Developed SLSSA by integrating four distinct search strategies, including a multiple food sources strategy.
• Implemented a self-learning mechanism to dynamically adjust the probability of executing each search strategy based on solution quality.
• Proposed an automated parameter setting method to optimize SLSSA performance without trial-and-error.

Main Results:

• SLSSA demonstrated superior performance against state-of-the-art algorithms on CEC2014 benchmark functions.
• The algorithm achieved higher accuracy, stability, and faster convergence speeds in training multi-layer perceptron classifiers on UCI datasets.
• SLSSA showed significant performance gains with only a minor increase in computational time compared to the original SSA.

Conclusions:

• The proposed SLSSA effectively addresses the limitations of traditional swarm intelligence algorithms.
• SLSSA offers a robust and efficient optimization tool for diverse applications, including machine learning.
• The self-learning mechanism and automated parameter tuning significantly enhance optimization capabilities.