SS-MSDA: Streamlined Sample-level Multi-source Domain Adaptation for EEG Emotion Recognition.

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• Affective Brain-Computer Interfaces (BCI) show promise but struggle with accurate emotion recognition for new users.
• Existing Multi-Source Domain Adaptation (MSDA) methods for BCI have limitations in performance, preparation complexity, and theoretical grounding.

Purpose of the Study:

• To propose an innovative MSDA algorithm to address the challenges in affective EEG-based BCI emotion recognition.
• To theoretically constrain the emotion classification error by narrowing the Wasserstein Distance between subdomain and target domain.

Main Methods:

• Developed a novel MSDA algorithm focusing on Wasserstein Distance minimization.
• Evaluated the algorithm on the SEED emotional EEG dataset.

Main Results:

• Achieved a 1-14% increase in recognition accuracy (average 7.2%) across subjects compared to baseline models.
• Significantly reduced preparation time by over 99.8% with minimal computational costs.
• Demonstrated superior performance over existing Domain Adaptation (DA) benchmarks.

Conclusions:

• The proposed SS-MSDA algorithm offers a practical and effective solution for enhancing emotion recognition in affective BCIs for new subjects.
• The algorithm extends MSDA theory and provides a novel perspective on constraining classifier error bounds without pre-training or data pre-collection.