Variability of neural activity has been shown to play a role in a variety of computational processes, including attention, learning, movement variability, and many others. However, much is still unknown about the computational role of neural variability and the constraints placed on it. In this dissertation, we describe three projects that aim to advance our understanding of neural variability by studying its structure, components, and controllability. The dissertation includes the following three parts: First, we studied the structure of neural variability using network models and in vivo recordings. In this project we found that the dimensionality and the proportion of variance shared among neurons depends on factors including the number of recorded neurons, the number of observed time points, and the underlying network connectivity structure. Second, we describe how neural variability can be separated into a component shared across hemispheres and components shared among neurons within the same hemisphere. We then show that the across hemisphere activity contains signatures of widespread cognitive processes. Third, we describe our efforts to develop a non-motor brain machine interface. This work leads up to a brain computer interface paradigm in which a non-human primate subject is able to use neurofeedback to suppress neural variability in prefrontal cortex. Together these three projects describe neural variability as originating from a combination of local and multi-area processes that are at least partially under volitional control.
Byron Yu (Co-Chair)
Matthew Smith (Co-Chair, University of Pittsburgh)
Rui Costa (Columbia University)
Zoom Participation Enabled. See announcement.