Brain machine interfaces incorporating learned dynamical structure in the brain

Millions of people are unable to move due to neurological injury or disease. Brain-machine interfaces seek to restore lost motor function to patients suffering such neurological deficits. Stanford researchers have discovered a way to provide a new class of brain-machine interface (BMI) algorithms to significantly improve performance over existing algorithms. These novel algorithms utilize learned dynamical structure in the brain for BMIs. This dynamical structure assumes an underlying lower-dimensional and latent state in the brain ("neural state"), which is the state of a dynamical system. If the dynamical structure is present in the brain, then decoding algorithms can leverage said structure to improve the performance of a BMI.

Figure 1 - BMI seek to restore lost motor function

Figure 2 - An example graph theoretic implementation of a dynamical BMI. The BMI incorporates an underlying neural state, s_k, which obeys dynamics and generates both the kinematics of the prosthetics device (x_k) and the neural data observed (y_k).