Extreme Learning Machine With Subnetwork Hidden Nodes for Regression and Classification.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Neural Networks

Background:

• Single-hidden-layer feedforward neural networks exhibit slow learning speeds, hindering practical applications.
• Extreme Learning Machine (ELM) offers significant speed improvements over traditional methods.
• Existing ELM methods often require a large number of hidden nodes, impacting efficiency.

Purpose of the Study:

• To develop an ELM-based learning method that enhances learning speed and efficiency.
• To improve generalization performance while reducing the number of hidden nodes.
• To overcome the limitations of conventional neural network training.

Main Methods:

• Proposed an ELM variant that dynamically grows subnetwork hidden nodes.
• Utilized residual network error feedback to the hidden layer for node growth.
• Validated the method against backpropagation, support vector machines, and other ELM techniques.

Main Results:

• The proposed method achieves learning speeds hundreds of times faster than traditional ELMs, backpropagation, and support vector machines.
• Demonstrated similar or superior generalization performance with significantly fewer hidden nodes.
• Experimental validation conducted on 32 diverse datasets.

Conclusions:

• The novel ELM approach offers a substantial advancement in learning speed and efficiency.
• Reduced hidden node requirements lead to more compact and potentially more interpretable models.
• This method presents a promising alternative for applications bottlenecked by slow neural network training.