EEG-Oriented Self-Supervised Learning With Triple Information Pathways Network.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Deep learning for electroencephalogram (EEG) analysis is gaining traction for clinical monitoring and intention/emotion recognition.
• Existing methods often use limited perspectives, struggle with complex spectro-/spatiotemporal patterns, and exhibit high variability.

Purpose of the Study:

• To develop novel EEG-oriented self-supervised learning methods and a deep architecture for rich representation learning.
• To address intra-/inter-subject variability in EEG signals.
• To create a versatile deep learning framework applicable across different EEG paradigms without task-dependent architecture engineering.

Main Methods:

• Proposed novel EEG-oriented self-supervised learning techniques.
• Developed a novel deep architecture to integrate spectral, spatial, and temporal EEG signal characteristics.
• Implemented a feature normalization strategy to mitigate signal variability.
• Validated the framework on four public EEG datasets.

Main Results:

• The proposed deep learning framework effectively learns rich EEG representations.
• The method successfully addresses intra-/inter-subject variability.
• Achieved state-of-the-art performance compared to existing baselines across four diverse EEG datasets.
• Demonstrated the efficacy of a single network architecture for multiple EEG paradigms.

Conclusions:

• The novel self-supervised learning framework and deep architecture offer a robust approach to EEG analysis.
• The proposed methods enhance the accuracy and reliability of intention and emotion recognition from EEG.
• This work advances EEG-based applications by providing a generalized and high-performing deep learning solution.