From brain waves to words: a new path to communication without surgery
2 days ago
- #Non-Invasive Decoding
- #Brain-Computer Interface
- #AI Neuroscience
- Brain2Qwerty v2 is a non-invasive AI system that decodes brain activity into text in real-time, achieving high accuracy without surgical implants.
- The system was trained on 22,000 sentences from nine participants using MEG recordings and end-to-end deep learning, leveraging large language models for semantic context.
- It achieved a 61% word accuracy rate overall and 78% for the best participant, significantly outperforming other non-invasive methods (8% accuracy).
- Decoding accuracy improves log-linearly with data volume, suggesting potential to close the gap with invasive techniques through scaling.
- Full training code for v1 and v2 is released, along with the v1 dataset from BCBL, to accelerate neuroscience and help people with communication impairments.
- The work is part of broader efforts to build open foundational brain models and includes collaborations like the $5 million Digital Brain Project fund.