Conditional Domain Adaptation Based on Initial Distribution Discrepancy for EEG Emotion Recognition

How to integrate data in different feature spaces and distributions is a research hotspot in EEG-based emotion recognition. A novel source-domain adaptation strategy based on initial distribution differences for EEG emotion recognition is proposed, which selects several source domains that are most similar to the target domain for domain adaptation. Compared to the ‘source-target pair’ domain adaptation method using all source domains, this method improves accuracy by up to 10% and reduces computation time by up to 43% based on the SEED-III and SEED-IV datasets.