MS-FRAN: A Novel Multi-Source Domain Adaptation Method for EEG-Based Emotion Recognition.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Electroencephalogram (EEG)-based emotion recognition is a growing research area.
• Significant challenges exist due to large distribution differences in EEG signals across subjects.
• Existing methods struggle to generalize effectively across different individuals.

Purpose of the Study:

• To propose a novel method, Multi-Source Feature Representation and Alignment Network (MS-FRAN), to overcome cross-subject variability in EEG emotion recognition.
• To enhance the alignment of EEG signal distributions between source and target domains.
• To reduce distributional differences among multiple source domains for improved model robustness.

Main Methods:

• Developed the Multi-Source Feature Representation and Alignment Network (MS-FRAN).
• Incorporated three key modules: Wide Feature Extractor (WFE) for feature learning, Random Matching Operation (RMO) for training, and Top-h ranked domain classifier selection (TOP) for classification.
• Focused on aligning distributions between source and target domains and reducing multi-source domain differences.

Main Results:

• MS-FRAN demonstrated effectiveness in aligning distributions between paired source and target domains.
• The method successfully reduced distributional differences among multiple source domains.
• Experimental results on SEED and DEAP datasets showed superior performance compared to existing approaches for cross-subject EEG emotion recognition.

Conclusions:

• The proposed MS-FRAN method offers a significant advancement in cross-subject EEG-based emotion recognition.
• Effective domain alignment and reduction of inter-subject variability are key contributions.
• The approach shows strong potential for real-world applications requiring robust emotion detection from EEG signals.