EVA-Net: Subject-Independent EEG Motor Decoding with Video-Derived Motor Priors
Researchers have developed EVA-Net, a novel two-stage framework designed to improve subject-independent EEG motor decoding for Brain-Computer Interfaces. This system leverages action videos as semantic priors to overcome the limitations of subject variability and signal non-stationarity that hinder current BCIs. By aligning EEG and video features in a shared space and then transferring knowledge from video prototypes to an EEG-only classifier, EVA-Net demonstrates significant accuracy gains on public datasets, outperforming text-based semantic anchors. AI
IMPACT This research could lead to more robust and user-friendly Brain-Computer Interfaces by improving the accuracy of decoding motor intentions from EEG signals.