Functional extreme learning machine.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Neural Networks

Background:

• Extreme Learning Machine (ELM) is a fast training algorithm for single hidden layer feedforward neural networks (SLFNs).
• ELM exhibits limitations including structure selection issues, overfitting, and suboptimal generalization performance.

Purpose of the Study:

• Propose a novel Functional Extreme Learning Machine (FELM) model.
• Utilize functional neurons and functional equation solving theory for improved model design.

Main Methods:

• Develop a new functional neuron (FN) model as the fundamental unit.
• Implement learning in FELM by adjusting basis function coefficients within neurons.
• Introduce a simple, iterative-free, high-precision, fast parameter learning algorithm.

Main Results:

• FELM learning adjusts basis function coefficients.
• A fast, iterative-free parameter learning algorithm was developed.
• Experimental results on UCI and StatLib datasets show FELM outperforms ELM and SVM in regression problems.

Conclusions:

• The proposed FELM model offers enhanced performance compared to existing algorithms like ELM and SVM.
• FELM provides a viable alternative for regression tasks, addressing ELM's shortcomings.