A new optimized GA-RBF neural network algorithm.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computational Neuroscience

Background:

• Radial Basis Function (RBF) neural networks offer adaptive and self-learning capabilities.
• Challenges with RBF networks include determining hidden layer neuron count and low weight learning efficiency, impacting performance.
• These limitations can decrease learning ability and recognition precision in complex problem-solving.

Purpose of the Study:

• To propose an optimized RBF neural network algorithm using a genetic algorithm (GA-RBF).
• To address the limitations of traditional RBF networks in determining network structure and optimizing weights.
• To enhance the learning ability and recognition precision of RBF networks for complex tasks.

Main Methods:

• Developed a novel Genetic Algorithm-RBF (GA-RBF) algorithm for optimizing RBF neural networks.
• Employed a hybrid encoding strategy: binary encoding for hidden layer neuron count and real encoding for connection weights.
• Optimized hidden layer neuron number and connection weights simultaneously, followed by Least Mean Square (LMS) algorithm refinement.

Main Results:

• The GA-RBF algorithm demonstrated improved operational efficiency when handling complex problems.
• Experimental results showed a significant enhancement in recognition precision compared to traditional methods.
• Validation on two UCI standard datasets confirmed the algorithm's effectiveness.

Conclusions:

• The proposed GA-RBF algorithm effectively optimizes RBF neural network structure and weights.
• The hybrid optimization approach leads to superior performance in complex problem-solving scenarios.
• The GA-RBF algorithm represents a valid and improved model for enhanced learning and recognition precision.