An ART-based construction of RBF networks

Area of Science:

• Artificial Intelligence
• Machine Learning
• Neural Networks

Background:

• Radial Basis Function (RBF) networks are common for function modeling.
• Traditional RBF networks face challenges with initial configuration and performance on abrupt or constant data.
• These limitations hinder their effectiveness in complex modeling tasks.

Purpose of the Study:

• To propose a novel approach to overcome the limitations of traditional RBF networks.
• To enhance the efficiency and accuracy of RBF network-based function modeling.
• To develop an adaptive method for RBF network construction.

Main Methods:

• Utilized new kernel functions for hidden nodes in RBF networks.
• Implemented an adaptive resonance theory (ART)-like algorithm for automatic node determination.
• Employed gradient-descent optimization for parameter and weight tuning.

Main Results:

• Constructed RBF networks with a reduced number of nodes.
• Achieved faster learning speeds compared to traditional methods.
• Demonstrated smaller approximation errors, indicating improved performance.

Conclusions:

• The proposed method offers a more efficient and accurate approach to RBF network construction.
• Automatic node determination and novel kernel functions address key limitations of TRBFs.
• This advancement is beneficial for function modeling applications requiring high precision and speed.