Self-Organizing RBF Neural Network Using an Adaptive Gradient Multiobjective Particle Swarm Optimization.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computational Neuroscience

Background:

• Radial basis function (RBF) neural networks often converge to local minima, hindering optimal performance.
• Optimizing RBF network structure and parameters (centers, widths, weights) is crucial for effective application.
• Existing methods struggle to balance network accuracy with complexity.

Purpose of the Study:

• To develop an adaptive gradient multiobjective particle swarm optimization (AGMOPSO) algorithm for RBF neural network optimization.
• To create an AGMOPSO-based self-organizing RBF neural network (AGMOPSO-SORBF) capable of optimizing network size and parameters.
• To achieve a balance between RBF network accuracy and structural complexity.

Main Methods:

• Designed an AGMOPSO algorithm incorporating a multiobjective gradient method and self-adaptive flight parameters.
• Developed the AGMOPSO-SORBF model to optimize RBF network parameters and determine network size.
• Performed convergence analysis for AGMOPSO-SORBF to ensure application reliability.

Main Results:

• The AGMOPSO-SORBF demonstrated superior generalization capability compared to existing methods.
• The proposed approach resulted in more compact RBF neural network structures.
• Numerical examples validated the effectiveness of the AGMOPSO-SORBF method.

Conclusions:

• The AGMOPSO-SORBF effectively addresses the local minima problem in RBF neural networks.
• This novel approach offers improved performance and efficiency in RBF network design.
• The AGMOPSO-SORBF provides a promising solution for developing accurate and compact RBF networks.