Self-Supervised Contrastive Pre-Training for EEG-Based Recognition via Cross Device Representation Consistency.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Electroencephalography (EEG) is vital for brain state modeling but faces challenges like low signal-to-noise ratios and scarce labeled data.
• Existing methods often address these challenges independently, limiting overall effectiveness.

Purpose of the Study:

• To introduce a novel pretraining paradigm, Cross-Device Representation Consistency (CDRC), for EEG recognition systems.
• To simultaneously address low signal-to-noise ratios and data scarcity in EEG analysis.

Main Methods:

• Developed CDRC, a self-supervised pretraining approach using representation distances and contrastive estimation.
• Employed a transformer-based dual-branch architecture with a contrastive feature alignment module.
• Evaluated on emotion classification (low SNR, dry electrodes) and vigilance regression (multimodal fusion, cross-device).

Main Results:

• CDRC achieved performance comparable to fully supervised methods on emotion classification and vigilance regression tasks.
• Reached state-of-the-art results among existing self-supervised methods, setting a new benchmark.
• Demonstrated strong performance on subject-independent tasks, effectively mitigating subject variability.

Conclusions:

• CDRC significantly enhances the practicality and scalability of EEG-based recognition systems.
• The method shows great potential for real-world brain-computer interfaces.
• CDRC offers a robust solution for improving EEG data analysis under challenging conditions.