Statistical Physics of Machine Learning

Explore fundamental questions at the intersection of statistical physics and machine learning in this presidential lecture that delves into recent trends and theoretical advances. Gain insights into how diffusion and flow-based generative models handle complex probability distributions, understand the phase transition between positional and semantic learning through a dot-product attention model, and examine classical uncertainty estimation methods in the context of modern overparameterized neural networks. Learn from physicist Lenka Zdeborová as she bridges the gap between natural sciences and machine learning, demonstrating how theoretical physics can help answer crucial questions about machine learning systems and their applications.