Quantized minimum error entropy with fiducial points for robust regression.

Area of Science:

• Machine Learning
• Signal Processing
• Statistical Modeling

Background:

• Minimum Error Entropy with Fiducial Points (MEEF) effectively mitigates non-Gaussian noise.
• MEEF's computational cost is high due to double summation over all error samples.
• Efficient noise reduction methods are crucial for robust data analysis.

Purpose of the Study:

• To develop a computationally efficient variant of MEEF.
• To introduce Quantized Minimum Error Entropy with Fiducial Points (QMEEF).
• To apply QMEEF for training linear models on noisy datasets.

Main Methods:

• An efficient quantization method is employed to represent error samples with a smaller subset.
• Theoretical properties of the proposed QMEEF are presented and proven.
• QMEEF is utilized to train linear regression, random vector functional link networks, and broad learning systems.

Main Results:

• The proposed QMEEF significantly reduces computational burden compared to MEEF.
• QMEEF demonstrates desirable performance in regression tasks with contaminated data.
• Experimental results validate the effectiveness of QMEEF on various datasets.

Conclusions:

• QMEEF offers an efficient and effective approach for handling non-Gaussian noise in machine learning and signal processing.
• The method provides a practical solution for training linear models with noisy data.
• QMEEF represents a valuable advancement in robust statistical modeling.