Brain2Qwerty v2 – Meta's non-invasive brain-computer interface
10 hours ago
- #brain-decoding
- #neurotechnology
- #MEG
- Brain2Qwerty v2 decodes natural sentences from non-invasive MEG recordings without requiring timing of keypresses.
- The model uses hierarchical modules to jointly improve letter, word, and sentence decoding, achieving up to 78% word accuracy.
- Two major challenges remain: performance needs improvement for practical use, and MEG devices must become more accessible and wearable.