Neurally Coupled Subspace Learning of Movement

Neural decoding of motor control of hand and arm movements in primates is a challenging task that requires developing statistical models that explain how the recorded neural population activity relates to motor behavior. Until recently, much of the work in this area has focused on learning linear models of decoding for low-dimensional motor control, such as 2D control of a computer cursor. Capturing a richer set of motor behaviors such as hand and arm posture during object grasping and manipulation tasks introduces much higher dimensional representations of motor control. Understanding the underlying degrees of freedom in complex kinematics that are explained by the neural activity is a central question. One way of learning these “effective” degrees of freedom has been to employ dimensionality reduction techniques, such as Principal Component Analysis, to find a linear kinematic subspace that accounts for the observed motor behavior, separate from the observed neural activity. The orthonormal bases that span this subspace are then considered as the underlying latent variables, or “motor primitives” that describe behavior. These motor primitives are not guaranteed to be optimally correlated with the observed neural activity however. In this paper we devise an objective function and optimize it to learn a linear subspace of the motor activity that tries to maximize the correlation between the latent variables of this subspace and the neural activity, while still explaining the motor behavior with reasonable fidelity.