One pass learning for generalized classifier neural network.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Computational Neuroscience

Background:

• Generalized classifier neural networks (GCNNs), a type of radial basis function network, use gradient descent for smoothing parameter optimization.
• This optimization process is time-consuming and can be a significant drawback in practical applications.

Purpose of the Study:

• To introduce a novel one-pass learning algorithm for GCNNs to address the computational inefficiency of traditional optimization methods.
• To enhance classification speed and maintain accuracy in GCNNs.

Main Methods:

• The proposed method calculates smoothing parameters using the standard deviation of each class.
• Two distinct functions are defined for smoothing parameter calculation, with thresholding to select the appropriate function based on dataset characteristics (e.g., range of values, standard deviation).
• The method was tested on 14 datasets and compared against probabilistic neural networks, RBF networks, ELMs, and standard/logarithmic learning GCNNs.

Main Results:

• The one-pass learning GCNN achieved classification speeds over a thousand times faster than standard and logarithmic learning GCNNs.
• The proposed method demonstrated comparable classification accuracy to other leading methods.
• The approach effectively overcomes the computational drawbacks of standard GCNN optimization.

Conclusions:

• One-pass learning GCNN offers a computationally efficient and accurate alternative for classification tasks.
• The method's speed and accuracy make it a viable replacement for probabilistic neural networks in certain applications.
• This approach enhances the overall performance and practicality of generalized classifier neural networks.