Extreme learning machines for regression based on V-matrix method.

Area of Science:

• Machine Learning
• Statistical Inference
• Data Mining

Background:

• Extreme Learning Machines (ELM) are efficient feedforward neural networks.
• V-matrix offers a novel framework for statistical inferences.
• Regression problems require robust and accurate predictive models.

Purpose of the Study:

• To investigate the synergistic effect of V-matrix and ELM on regression tasks.
• To develop efficient algorithms for V-matrix evaluation and weighted ELM.
• To address potential biases in V-matrix weighting for improved regression performance.

Main Methods:

• A novel algorithm for efficient V-matrix evaluation.
• Development of V-ELM, a weighted ELM algorithm utilizing V-matrix.
• Introduction of VI-ELM to mitigate V-matrix bias by minimizing regression and V-matrix weighted errors simultaneously.

Main Results:

• The proposed algorithms, V-ELM and VI-ELM, demonstrate effectiveness in regression problems.
• VI-ELM successfully addresses the bias of assigning higher weights to instances with smaller input values.
• Experimental validation on 12 real-world benchmark datasets confirms the proposed methods' efficacy.

Conclusions:

• The integration of V-matrix with ELM offers a promising approach for regression.
• VI-ELM provides a robust solution by balancing regression accuracy and V-matrix weighting.
• The developed methods enhance the capability of ELM in handling complex regression challenges.