Neural decoding with co-learning for brain computer interfaces

Abstract

Embodiments relate to decoding architecture that rapidly decodes light-derived signals to extract predicted user actions or intents (e.g., commands) in relation to interactions with objects (e.g., virtual objects, physical objects), such that the user can manipulate the objects or otherwise receive assistance without manually interacting with an input device (e.g., touch input device, audio input device, etc.). The decoding architecture thus enables a neural decoding process with a neural signal stream as an input, and provides feedback to the user, where the feedback is used to train the neural decoding algorithm and user behavior. The neural signals can be blood oxygenation level dependent (BOLD) signals associated with activation of different articulators of the motor cortex, and signals can characterize both actual and imagined motor cortex-related behaviors. With training of the decoding algorithm, rapid calibration of the BCI for new users can be achieved.