How self-supervised learning can help us build better representations of neural activity

The Duke CTN group welcomes Eva Dyer, Phd, assistant professor of biomedical engineering at Georgia Tech, for her seminar "How self-supervised learning can help us build better representations of neural activity". The event is on Zoom; please contact d.shipman@duke.edu for connection details. ABSTRACT: In the brain, representations of stimuli, intent, and behavior are formed, both locally at the level of neural circuits and also over larger distributed circuits spanning distinct brain areas. Being able to understand how neural circuits, both at the local and macroscale, coordinate to drive behavior and decision making, is a fundamental goal of neuroscience. In this talk, I will describe recent work from my lab that aims to leverage principles underlying 'self-supervised learning' to learn more robust and interpretable representations of neural population activity. With new tools to read out information from the brain in a stable and generalizable manner, we can enable comparisons across neural recordings spanning different time points in learning, or across aging and development.