Learning aligned EEG representations with subject-specific encoders

The study presents a novel approach to EEG decoding by implementing subject-specific encoders in place of a shared encoder, followed by a common classifier, and compares this hybrid model against baselines like EEGNet, AttentionBaseNet, and CTNet with Euclidean Alignment (EA). The findings indicate that while EA enhances shared encoders, the hybrid model effectively learns subject-aligned representations, improving class distinctiveness and reducing inter-subject variability. This research highlights the potential of subject-specific encoders as an effective alignment mechanism in EEG decoding, while also identifying challenges in generalizing to unseen subjects.