Instance-representation transfer method based on joint distribution and deep adaptation for EEG emotion recognition.

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• Electroencephalogram (EEG) based emotion recognition is crucial for advancing human-computer interaction.
• Practical applications are hindered by subject and session variability.
• Transfer learning offers a solution, with existing methods focusing on instance or representation transfer.

Purpose of the Study:

• To propose a novel emotion recognition method, Joint Distributed Instances Represent Transfer (JD-IRT), to address EEG variability.
• To enhance the accuracy and applicability of EEG-based emotion recognition systems.

Main Methods:

• Developed JD-IRT, comprising Joint Distribution Deep Adaptation (JDDA) and Instance-Representation Transfer (I-RT).
• JDDA addresses marginal and conditional distribution discrepancies using adaptive layers and kernels for deep domain adaptation.
• I-RT employs instance transfer to optimize source domain data selection for improved representation transfer.

Main Results:

• Achieved cross-subject accuracies of 83.21% (SEED), 52.12% (SEED-IV), and 60.17% (SEED-V).
• Achieved cross-session accuracies of 91.29% (SEED), 59.02% (SEED-IV), and 65.91% (SEED-V).
• Demonstrated significant improvements over existing representative methods on benchmark datasets.

Conclusions:

• The proposed JD-IRT method effectively improves EEG emotion recognition accuracy across different subjects and sessions.
• This advancement holds promise for more robust and practical human-computer interaction systems.
• The study highlights the efficacy of combining deep domain adaptation with instance transfer for cross-domain learning in EEG analysis.