Global Adaptive Transformer for Cross-Subject Enhanced EEG Classification.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Decoding electroencephalography (EEG) signals for mental intention recognition is challenging due to significant inter-subject variability.
• Existing transfer learning methods for cross-subject EEG analysis often struggle with inadequate feature representation and fail to capture long-range dependencies.

Purpose of the Study:

• To propose a novel domain adaptation method, Global Adaptive Transformer (GAT), to leverage source subject EEG data for enhancing target subject's mental intention decoding.
• To address limitations in feature representation and long-range dependency capture in current cross-subject EEG analysis.

Main Methods:

• GAT utilizes parallel convolution for initial temporal and spatial feature extraction from EEG signals.
• An attention-based adaptor implicitly transfers source domain features to the target domain, focusing on global EEG feature correlations.
• A discriminator and adaptive center loss are employed to reduce marginal and align conditional distribution discrepancies between source and target domains, respectively.

Main Results:

• GAT significantly outperforms state-of-the-art methods on two benchmark EEG datasets.
• The proposed attention-based adaptor is identified as a key component contributing to the method's superior performance.
• The study demonstrates the effectiveness of GAT in improving cross-subject EEG signal decoding.

Conclusions:

• Global Adaptive Transformer (GAT) offers a robust solution for cross-subject EEG analysis in brain-computer interfaces (BCI).
• The method shows strong potential for enhancing the practical applicability of BCI systems by improving the decoding of mental intentions.
• GAT's ability to effectively adapt features across subjects opens new avenues for personalized BCI development.