Regan Zeaser

Major: Integrative Neuroscience

Biography: Regan is a senior majoring in Integrative Neuroscience with a Systems concentration. She conducted her neuroscience capstone with Dr. Stephen Keeley, and her research investigated the ability of a Generalized Linear Model to infer neural connectivity in real neural data. In the future, Regan plans to attend graduate school and continue pursuing her research interests in Machine Learning and Neuroscience.

Project Title: Fitting a Generalized Linear Model to a Novel Neural Dataset

Faculty Mentor: Stephen Keeley, Department of Natural Sciences

Abstract: The Generalized Linear Model (GLM) is a Machine Learning algorithm that can be used to analyze neural data by relating inputs, whether they be from an external stimulus or another neuron, to ‘outputs’, or spiking of the neuron(s). GLMs have become a popular way to predict neuronal behavior due to their tractability and their ability to reflect the dynamic activity of a neuron. In this study, a maximum likelihood estimation procedure was used to learn the parameters of a Linear-Nonlinear Poisson GLM and to infer connectivity from raw spike trains from multiple neurons. This model includes a linear map between parameters and stimuli, followed by a soft-plus non-linearity to give rise to a Poisson-rate parameter lambda. We first validate the approach by simulating spike trains with known connectivity and are able to recover true parameters. We then test the model on real recorded spike trains from rodent striatal neurons in an evidence accumulation task.