Machine Learning in BCI

Applying modern pattern recognition architectures to decode complex, categorical intentions from rich neural feature spaces.

Key takeaway: While Kalman Filters reign supreme for decoding continuous states tied to physics (like smoothing the velocity of a cursor), Machine Learning algorithms (from simple SVMs to deep RNNs) are the state-of-the-art for decoding distinct, categorical intents. They are the engine behind predicting which specific phoneme a paralyzed patient is trying to speak based on the complex spatiotemporal patterns of their cortical activity.

Classical Machine Learning

Deep Learning Frameworks

The Grand Challenge: The Data Bottleneck