Transfer Discriminative Dictionary Pair Learning Approach for Across-Subject EEG Emotion Classification.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Electroencephalogram (EEG) signals are valuable for noninvasive emotion recognition.
• Individual differences in EEG data pose challenges for cross-subject emotion classification.
• Collecting large labeled datasets for new subjects is often impractical for traditional methods.

Purpose of the Study:

• To propose a transfer discriminative dictionary pair learning (TDDPL) approach for robust across-subject EEG emotion classification.
• To develop a model that generalizes well to new subjects without requiring extensive subject-specific labeled data.
• To overcome the limitations of individual differences in EEG signal distributions.

Main Methods:

• The TDDPL approach projects data into a domain-invariant subspace.
• It utilizes a maximum mean discrepancy (MMD) strategy to build a transfer dictionary pair.
• Shared synthesis and analysis dictionaries are learned to transfer discriminative knowledge from source to target domains.

Main Results:

• The TDDPL approach effectively minimizes reconstruction error and maximizes inter-class separation.
• It learns discriminative synthesis dictionaries and sparse, low-rank coding.
• A discriminative classifier is constructed in the target domain without calculating coding coefficients.

Conclusions:

• The proposed TDDPL method demonstrates effectiveness for across-subject EEG emotion classification.
• This approach addresses the challenge of individual differences in EEG data.
• TDDPL offers a promising solution for building generalizable emotion recognition models.