I am a Research Associate in the Psychology department at the University of Pennsylvania (Philadelphia, PA). I am researching how the brain represents learning, and how learning circuits can be modeled using machine learning algorithms. My undergraduate education was in Psychobiology at Florida Atlantic University’s Center for Complex Systems and Brain Sciences. In 2002 I moved to Winston Salem, NC to pursue a Neuroscience PhD at Wake Forest Medical Center. There I studied how supervised learning (working memory) tasks altered the functional organization of spatial and shape representation in the prefrontal cortex. In August 2008 I joined Carl Olson's laboratory as a postdoctoral fellow at Carnegie Mellon University to study visual object recognition and learning in inferotemporal cortex. I created experimental designs to examine the neural circuits of unsupervised learning in monkeys. In 2015 I began working with Nicole Rust to study learning at the neural population level using machine learning algorithms.

Research Overview

The brain is amazing in its ability to process incredible amounts of information in a short time. Indeed, much of the recent breakthroughs in machine learning and artificial intelligence has come from modeling neural circuits. My research focuses on understanding how the brain learns to processes large amounts of streaming data. I have developed a number of experimental designs to explore supervised and unsupervised learning in the visual system of human and non-human primates (See Research). One of the interesting findings was that the brain automatically learns the statistical relationships of environmental stimuli by making predictions, and suppressing the neural activity for that accurately predicted events. In a fully predictable environment, this would lead to a 'Zen-like' state where the brain only requires enough sensory input to test the internal prediction. In other words, through experience we learn what events are likely to occur, and if those predictions are accurate the brain does not need to fully process those images. This can conserve vital computational, and metabolic resources.