Multi-subject subspace alignment for non-stationary EEG-based emotion recognition.

Area of Science:

• Neuroscience
• Machine Learning
• Human-Computer Interaction

Background:

• Emotion recognition using electroencephalogram (EEG) signals is vital for human-machine collaboration and psychiatric diagnostics.
• EEG signal variability across subjects (due to fatigue, electrode placement, impedance) necessitates time-consuming individual calibration, limiting practical application.
• Existing domain adaptation (DA) techniques primarily address single-subject to single-subject adaptation, restricting their real-world utility.

Purpose of the Study:

• To develop a novel unsupervised domain adaptation strategy for robust EEG-based emotion recognition across multiple subjects.
• To create personalized emotion recognition models without requiring user-specific labeled data.
• To overcome the limitations of existing DA methods in handling multi-subject scenarios.

Main Methods:

• Proposed a novel unsupervised domain adaptation strategy named Multi-Subject Subspace Alignment (MSSA).
• Integrated subspace alignment solutions with multi-subject information within a unified framework.
• Validated the approach using the public SEED EEG dataset.

Main Results:

• MSSA demonstrated effectiveness in building personalized EEG emotion recognition models without labeled user data.
• The proposed MSSA method outperformed state-of-the-art techniques in multi-subject adaptation scenarios.
• Achieved superior performance in cross-subject emotion recognition tasks.

Conclusions:

• MSSA offers a significant advancement in unsupervised domain adaptation for EEG-based emotion recognition.
• The method successfully addresses the challenge of inter-subject variability, enabling more practical applications.
• MSSA provides a robust and efficient solution for personalized emotion recognition systems.