CIF: Medium: Collaborative Research: Scalable Learning of Nonlinear Models in Large Neural Populations

New York University, New York NY

Investigators

Abstract

Fundamental to understanding information processing in the brain are methods that can systematically characterize the structure and dynamics of neural circuits that underlie perception and cognition. Micro- electrocorticography (µECoG) is the practice of using microelectrodes placed directly on the exposed surface of the brain to record electrical activity from the cerebral cortex. Recent advances in µECoG provide unique opportunities to observe large regions of the neural cortex at unprecedented spatial and temporal resolution. However, uncovering the structure of complex neural circuits is challenging. This interdisciplinary project develops methods for learning high-dimensional nonlinear systems with a particular focus on these systems as they arise in cortical networks and validates these techniques on state-of-the-art µECoG systems. Three thrusts are considered: The first considers the general problem of state estimation in high-dimensional dynamical systems using decomposition methods including distributed Kalman and particle filtering and graphical models. The main goal is to provide computationally scalable and flexible approaches with provable guarantees. The second combines these state estimation methods with Bayesian parameter estimation and compressed sensing techniques to identify connectivity and nonlinear dynamics in the networks. The third validates these methods on identification of neural models from µECoG arrays. Applications to neural mapping, auditory and visual stimuli decoding are explored. In particular, the project seeks to demonstrate the method on using recordings from rat primary auditory cortex and cat visual cortex using a novel, flexible, high-resolution electrode array.