PRISM Boosts Cross-Subject EEG Emotion Recognition with Limited Data

Electroencephalogram (EEG) signals hold significant potential for accurately decoding human emotions due to their high temporal fidelity. However, two major hurdles prevent widespread adoption: the presence of redundant channels in EEG data and the substantial variability in brain activity patterns across different individuals. These issues make it difficult to develop models that generalize well from one subject to another. Researchers have introduced PRISM (PRioritized channel Importance with Semi-supervised doMain adaptation) to overcome these challenges. PRISM employs a two-pronged approach: it assigns data-dependent weights to EEG channels, effectively prioritizing relevant electrodes and suppressing noise, and it utilizes unlabeled data through confidence-filtered pseudo-labels to align different subjects' data, thereby reducing inter-subject heterogeneity. Extensive testing on multiple datasets shows PRISM significantly outperforms existing methods, achieving strong cross-subject generalization even with minimal labeled data.