Mehrdad Jazayeri, Ph.D. (Assistant Professor, Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology) will be presenting a talk titled, “A System Identification Approach to Infer Neural Codes from Neural Dynamics” at the Center for Sensorimotor Neural Engineering, Tuesday, October 11, 2016, 3:30 pm.
Summary: Recent advances in systems neuroscience have motivated a shift of perspective on neural code from static single-neuron codes such as tuning functions to codes that are governed by structure and dynamics of population activity. This shift of perspective invites a change of approach: instead of charting average firing rates of single neurons, we must infer behaviorally relevant neural codes from evolving activity patterns across neural populations. Here, Dr. Jazayeri will present a novel approach consisting of 3 inter-related steps to tackle this problem: 1) use high-density recording to characterize the intrinsic manifold along which neural dynamics evolve, 2) develop recurrent network models to generate hypotheses about the principles that control the dynamics, and 3) apply model-based perturbations to infer the key features of the observed dynamics that provide a predictive neural code for the behavior. Dr. Jazayeri will develop these ideas in the context of a project where we have found a candidate neural code for motor timing based on the neural dynamics in the medial frontal cortex.