EVA-Net: Subject-Independent EEG Motor Decoding with Video-Derived Motor Priors

Researchers propose EVA-Net, a machine learning framework that uses video-based motor priors to improve EEG brain-computer interfaces (BCIs) across different subjects with minimal calibration. The two-stage approach achieves 8.66% accuracy improvement over existing methods, demonstrating that video is a more effective semantic anchor than text for decoding motor intent from brain signals.

EVA-Net addresses a fundamental challenge in non-invasive brain-computer interface technology: the inability of current EEG decoders to generalize across different subjects without extensive individual calibration. This research tackles inter-subject variability—the biological differences in how individuals' brains produce electrical signals—by leveraging multimodal learning with video as a semantic foundation. The framework's two-stage design first aligns EEG and video features in a shared representational space, then transfers learned video knowledge to an EEG-only classifier, eliminating computational overhead during deployment.

The work builds on growing recognition that multimodal approaches enhance BCI performance, but importantly pivots from text-based supervision to dynamic video priors. This distinction matters because motor processes are inherently temporal and spatial; videos capture this dynamism better than static text descriptions. The 8.66% accuracy improvement on the EEGMMI dataset and strong cross-subject generalization suggest video-grounded learning more effectively separates true motor semantics from subject-specific neural noise.

For the BCI and neurotechnology sectors, improved subject-independent decoding directly addresses commercialization barriers. Current BCIs require individualized training sessions, limiting accessibility and real-world deployment. EVA-Net's approach could enable plug-and-play BCI systems requiring minimal calibration, accelerating adoption in medical rehabilitation, assistive technology, and human-computer interaction applications. The demonstrated superiority of video over text also provides a template for future multimodal BCI research.

Future development should focus on validating EVA-Net across larger, more diverse populations and exploring whether video priors transfer to non-motor cognitive tasks. Real-world robustness testing under varying conditions and integration with practical BCI hardware pipelines remain critical next steps for clinical viability.

• →EVA-Net uses video-derived motor priors to improve subject-independent EEG decoding, achieving 8.66% accuracy gains on standard benchmarks
• →Two-stage framework combines cross-modal alignment and knowledge distillation without adding inference computational cost
• →Video provides significantly more effective semantic supervision than text for decoding dynamic motor processes from brain signals
• →Strong cross-subject generalization reduces calibration requirements, addressing a major barrier to practical BCI commercialization
• →Results suggest multimodal learning with dynamic priors is a promising direction for generalizable brain-computer interfaces