Improving the Generalization Properties of Radial Basis Function Neural Networks.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computational Neuroscience

Background:

• Radial basis function neural networks (RBFNNs) offer fast, linear learning for complex nonlinear mappings.
• Effective generalization in RBFNNs depends on the smoothness of the network's mapping.
• Existing methods may not balance training speed with generalization performance.

Purpose of the Study:

• To introduce explicit smoothing into RBFNNs.
• To maintain fast training speeds while enhancing generalization.
• To demonstrate the effectiveness of a modified error functional.

Main Methods:

• Modification of the error functional to incorporate a smoothing term.
• Training RBFNNs with the modified error functional.
• Evaluation of generalization performance using a specific example.

Main Results:

• The modified error functional successfully introduced explicit smoothing.
• Generalization properties of the RBFNN were significantly improved.
• The modification did not substantially increase training time.

Conclusions:

• Explicit smoothing via error functional modification is an effective strategy for improving RBFNN generalization.
• This approach offers a practical way to enhance network performance without compromising computational efficiency.
• The findings suggest a valuable technique for developing more robust neural network models.