speaker: Francesca Mignacco, Princeton University
Abstract: Modern machine learning has achieved remarkable success across science and technology. However, progress has remained largely empirical, and a unified theory explaining how and why these systems work is lacking. In parallel, experimental neuroscience has enabled large-scale recordings of brain activity, calling for quantitative theories that link microscopic neural interactions to macroscopic behavior. This talk uses statistical physics as a unifying language to address these challenges through data-driven yet analytically tractable models. These models allow us to identify interpretable summary statistics that capture collective neural computations. I present examples of applications of these methods to study the inductive biases of neural architectures, the structure of real datasets, and the learning curves of training algorithms considered as controlled high-dimensional dynamic processes.
biography: Francesca Mignacco is a postdoctoral fellow at the Center for Biofunctional Physics, a collaboration between Princeton University and the City University of New York. Her research is at the intersection of statistical physics, machine learning, and computational neuroscience. She develops principle-based methods and models to reveal the low-dimensional structure of neural populations and investigate the mechanisms underlying neurodynamics and meta-learning.
host: Professor Ernst Witt
