Multi-source joint domain adaptation for cross-subject and cross-session emotion recognition from...

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• Affective brain-computer interfaces (BCIs) are crucial for advancing human-computer interaction.
• Emotion recognition using electroencephalography (EEG) faces challenges due to data distribution shifts across subjects and time.
• Existing domain adaptation methods often overlook joint distribution differences between multiple source domains.

Purpose of the Study:

• To propose a novel multi-source joint domain adaptation (MSJDA) network for robust EEG-based emotion recognition.
• To address the limitations of existing methods in handling varied data distributions in EEG.
• To improve the generalization of emotion recognition models across different subjects and sessions.

Main Methods:

• Developed a multi-source joint domain adaptation (MSJDA) network.
• Mapped multiple source domains to a shared feature space.
• Aligned joint distributions of private representations and classification predictions between source and target domains.

Main Results:

• The proposed MSJDA network demonstrated significant effectiveness on the SEED dataset.
• Achieved superior classification results, particularly in the challenging cross-subject emotion recognition task.
• Outperformed existing methods by better matching source and target domain distributions.

Conclusions:

• The MSJDA network offers a promising approach for enhancing EEG emotion recognition.
• Effective alignment of joint distributions is key to overcoming data variability in affective BCIs.
• The model shows strong potential for real-world applications requiring reliable emotion detection from EEG signals.