spectral-features

This paper shows that traditional machine learning models using spectral features from EEG signals can match or outperform state-of-the-art attention-based deep learning models for diagnosing neurodegenerative diseases, suggesting fundamental limitations of attention mechanisms in this domain.