Recognition of MI-EEG signals using extended-LSR-based inductive transfer learning.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Motor imagery electroencephalography (MI-EEG) signal recognition is crucial for brain-computer interface (BCI) systems.
• Current BCI systems often require extensive subject-specific labeled data for classification algorithms.
• Existing transfer learning methods for EEG signal recognition can be model-specific, limiting broad application.

Purpose of the Study:

• To develop a more broadly applicable transfer learning method for MI-EEG signal recognition.
• To address the challenge of insufficient subject-specific training data in BCI systems.
• To enhance the generalization capabilities of intelligent models used in BCI.

Main Methods:

• An extended-LSR-based inductive transfer learning method was proposed.
• The method facilitates transfer learning across diverse intelligent models, including neural networks, Takagi-Sugeno-Kang (TSK) fuzzy systems, and kernel methods.
• This approach aims to transfer knowledge from source domains to improve target domain performance with limited data.

Main Results:

• The proposed method effectively transfers knowledge from source domains to enhance learning performance in target domains with insufficient data.
• Incorporating multiple classic base models improved the application and generalization of the transfer learning method.
• Experimental results validated the method's effectiveness in MI-EEG signal recognition.

Conclusions:

• The developed extended-LSR transfer learning method offers a robust solution for MI-EEG signal recognition.
• This approach broadens the applicability and generalization of intelligent models in BCI systems.
• The method shows significant potential for improving BCI performance, especially when training data is scarce.