MS-MDA

MS-MDA applies a multi-branch domain adaptation strategy to EEG-based emotion recognition to mitigate variability between subjects and sessions.

Key Features:

• Multi-branch architecture: Constructs independent branches for each EEG data source domain to enable one-to-one domain adaptation.
• One-to-one domain adaptation: Adapts each source domain individually rather than treating diverse EEG datasets as a single source domain.
• Domain-specific feature extraction: Extracts domain-specific features from each branch while leveraging shared low-level features across sources.
• Handling subject and session variability: Manages variability across subjects and sessions by integrating diverse EEG data characteristics across branches.
• Adaptive inference integration: Integrates shared and distinct attributes from multiple branches for adaptive emotion recognition.

Scientific Applications:

• Emotion recognition from EEG: Enables EEG-based emotion classification that accounts for inter-subject and inter-session variability.
• Cross-subject/session domain adaptation: Supports studies aiming to adapt models across different subjects and recording sessions in EEG datasets.
• Evaluation of domain adaptation approaches in EEG: Provides a framework to address marginal distribution mismatches encountered by traditional DA methods on EEG data.

Methodology:

Constructs independent branches for each EEG data source domain to facilitate one-to-one domain adaptation, assumes common low-level features across sources and extracts domain-specific features through the multi-branch framework, and uses multiple branches to manage and integrate diverse EEG characteristics across subjects and sessions for inference.