Brain-Computer Interfaces

Key takeaway: Calibration-free reached 68.8% ± 17.6%; adding about 43 seconds of subject data increased accuracy to 92.0% ± 5.2%.

Key takeaway: Most non-deep methods clustered at mean AUC 0.64-0.65 and significantly outperformed deep models (best deep AUC: 0.57).

Key takeaway: Optimal channel subsets improved accuracy by 10-15%, and all tested methods completed in under 80 seconds.