A fast kernel extreme learning machine based on conjugate gradient.

Area of Science:

• Computational intelligence
• Machine learning
• Artificial neural networks

Background:

• Kernel Extreme Learning Machine (KELM) enhances Extreme Learning Machine (ELM) generalization and stability.
• KELM's performance is sensitive to a randomly set penalty parameter.
• A need exists for improved KELM methods with better parameter handling.

Purpose of the Study:

• Introduce a fast KELM variant using the conjugate gradient method (CG-KELM).
• Eliminate the penalty parameter in KELM.
• Enhance learning speed and generalization ability.

Main Methods:

• Developed CG-KELM by integrating the conjugate gradient method for output weight computation.
• Employed iterative conjugate gradient method, removing the need for manual penalty parameter tuning.
• Validated performance through image restoration simulations.

Main Results:

• CG-KELM demonstrates superior performance compared to standard KELM in image restoration tasks.
• The proposed method achieves faster learning speeds.
• CG-KELM offers improved generalization ability without a penalty parameter.

Conclusions:

• CG-KELM provides a robust and efficient alternative to KELM.
• The method balances the strengths of ELM and KELM.
• CG-KELM is a promising approach for machine learning applications requiring high generalization and speed.