A Cross-Attention-Based Class Alignment Network for Cross-Subject EEG Classification in a Heterogeneous Space

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Domain adaptation (DA) is crucial for cross-subject classification.
• Traditional DA methods assume homogeneous spaces, limiting real-world EEG classification.
• Heterogeneous space EEG classification remains a significant research challenge.

Purpose of the Study:

• To develop a novel framework for EEG classification under heterogeneous spaces.
• To address limitations of traditional DA methods in cross-domain scenarios.
• To improve cross-subject classification accuracy in complex, heterogeneous environments.

Main Methods:

• Implemented a cross-domain class alignment strategy.
• Developed a cross-encoder to capture inter-domain data dependencies.
• Introduced a tailored class discriminator with an optimizing loss function for feature aggregation and dispersion.

Main Results:

• The proposed method demonstrated superior performance on two public EEG datasets across five heterogeneous space scenarios.
• Outperformed advanced methods by an average of 7.8% in heterogeneous label space scenarios.
• Achieved an average improvement of 4.1% over state-of-the-art methods in complex heterogeneous feature and label space scenarios.

Conclusions:

• An efficient model for cross-subject EEG classification in heterogeneous spaces was presented.
• The framework effectively addresses challenges in heterogeneous EEG classification.
• Opens new research avenues in cross-domain learning for biomedical applications.