Tracking Hand Movements from Neuronal Activity with a Dynamic Kernel-Based Model
It is well known that population activity in motor cortex can predict movement direction. This allows for development of brain machine interfaces (BMI) that read brain activity and produce movements. BMI success depends heavily on the quality of data and on the efficiency of the algorithm used to interpret it. Here, we devise and experiment with a dynamical kernel-based system for tracking hand movements from neuronal activity. The state of the system corresponds to the hand location, velocity, and acceleration, while the system's inputs are the instantaneous spike rates. The system's state dynamics is defined as a combination of a linear mapping from the previous estimated state and a kernel-based mapping tailored for modeling neural activities. The state to state mapping encourages smooth movements with constant acceleration and minimal jerk. The mapping from neural activity to state is done using a kernel method that implicitly embeds the population's spike rate patterns into a high dimensional feature space. Learning consists of minimizing a convex, noise-robust loss function by weighing the effect of selected, previously seen, activity and movement examples. We use this approach to predict hand trajectories on the basis of neural activity of single units recorded simultaneously in motor cortex of behaving monkeys and find that the proposed approach is more accurate than both a static approach based on support vector regression and the Kalman filter.