Prototype-based Domain Generalization Framework for Subject-Independent Brain-Computer Interfaces.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Brain-computer interface (BCI) systems using electroencephalography (EEG) face challenges due to significant inter- and intra-subject variability.
• Current BCI systems often require frequent recalibration for each user and session, hindering practical application.

Purpose of the Study:

• To develop a subject-independent EEG classification framework that generalizes to unseen subjects without requiring additional calibration.
• To leverage domain generalization strategies, specifically open-set recognition, to enhance the performance of BCI systems on novel data.

Main Methods:

• Proposed a framework utilizing open-set recognition as an auxiliary task to learn subject-specific style features from source data.
• Employed a shared feature extractor designed to map features from unseen target datasets (new subjects) as distinct, unknown domains.
• Focused on achieving cross-instance style invariance within domains and minimizing open-space risk for improved generalization.

Main Results:

• Experiments demonstrated that incorporating domain information via an auxiliary network significantly enhances generalization performance.
• The proposed framework effectively maps features from unseen subjects, improving the robustness of the BCI system.
• The strategy showed potential in reducing the need for subject-specific calibration in EEG-based BCI.

Conclusions:

• The developed framework offers a promising strategy for improving subject-independent BCI performance.
• This approach can reduce the necessity for recalibration, making BCI systems more practical for real-world applications.
• The framework has potential applications in various mental state monitoring tasks, including neurofeedback, seizure detection, and sleep disorder analysis.