Computation & Interpretation

In recent years, many new computational methods have been created to make sense of the large amounts of data collected with high-density neural recording techniques. But many of these methods share critical limitations: they are under-constrained by neural data alone, they do not disambiguate the roles of different brain regions, and they do not generate the control signals necessary for movement. Because too many possible solutions exist, these methods struggle to predict inputs arriving from other brain areas or from sensory feedback.

To constrain model solutions and predict how brain areas interact, we develop frameworks for jointly modeling neural activity and skilled closed-loop motor control. These methods allow us to describe the contributions of different brain regions to control and to characterize the information communicated between them. This approach builds on our earlier finding that neural population activity during movement is better explained as an evolving dynamical system than by fixed representational tuning — a view of motor cortex that jointly modeling brain and behaviour now lets us take much further.