A Novel Transfer Support Matrix Machine for Motor Imagery-Based Brain Computer Interface.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Motor imagery (MI) brain-computer interfaces (BCIs) show promise but require extensive labeled electroencephalography (EEG) data.
• Subject-specific EEG variations necessitate lengthy calibration sessions for accurate model training.
• Transfer learning offers a potential solution by utilizing knowledge from reference subjects.

Purpose of the Study:

• To develop a novel Knowledge-Leverage Support Matrix Machine (KL-SMM) for enhancing MI-BCI classification performance.
• To address the challenge of limited labeled EEG data in target subjects.
• To leverage existing model knowledge from source domains while protecting privacy.

Main Methods:

• Developed a Knowledge-Leverage Support Matrix Machine (KL-SMM) capable of learning directly from matrix-form EEG data.
• Incorporated transfer learning to leverage both few target domain EEG data and source domain model knowledge.
• Optimized the KL-SMM objective function using the alternating direction method of multipliers (ADMM).

Main Results:

• The KL-SMM demonstrated superior classification performance compared to existing methods with insufficient EEG data.
• The model effectively leveraged limited target domain data and source domain knowledge.
• Experimental validation on public MI-based EEG datasets confirmed the KL-SMM's effectiveness.

Conclusions:

• The KL-SMM significantly improves generalization performance in MI-BCIs with limited labeled data.
• This approach reduces the need for extensive individual data collection, shortening calibration times.
• KL-SMM offers a privacy-preserving and efficient solution for subject-independent BCI applications.