Optimized Particle Swarm Optimization (OPSO) and its application to artificial neural network training.

Area of Science:

• Computational Intelligence
• Machine Learning
• Optimization Algorithms

Background:

• Particle Swarm Optimization (PSO) is a widely used parameter optimization technique.
• PSO's performance is sensitive to its strategy parameters, requiring careful selection for specific tasks.
• Meta-optimization of PSO parameters is essential for improving its convergence behavior.

Purpose of the Study:

• To introduce the Optimized Particle Swarm Optimization (OPSO) method for meta-optimizing PSO parameters.
• To evaluate OPSO's effectiveness on artificial fitness functions and compare it with existing PSO implementations.
• To apply OPSO to neural network training for quantitative prediction of blood-brain barrier permeation.

Main Methods:

• Developed OPSO, a meta-optimization technique where swarms optimize PSO's free parameters.
• Assessed OPSO performance against two popular PSO implementations using five artificial fitness functions.
• Applied OPSO to train a neural network model for predicting blood-brain barrier permeation.

Main Results:

• Meta-optimized parameters significantly improved PSO performance.
• OPSO enhanced optimization speed and quality compared to other PSO methods in most experiments.
• Neural network training time decreased by factors of four and two; a predictive model with good data correlation was achieved.

Conclusions:

• Optimizing PSO parameters through meta-optimization leads to performance gains.
• The OPSO approach provides effective parameter combinations for enhanced optimization.
• OPSO's straightforward implementation facilitates its practical application.