Abstract:

Generative AI has revolutionized how we model and synthesize data, powering applications such as realistic image and video generation. Yet, its foundations trace back to classical statistical methods. In this talk, we begin with a brief exploration of the evolution of generative modeling, from traditional probabilistic techniques to modern approaches like variational autoencoders and diffusion models. Beyond data generation, these models play a crucial role in learning and forecasting complex dynamical systems, with applications in physics, engineering, and beyond. We will discuss the surprising connection between optimal transport and a thought experiment of Schrödinger, and present a principled framework for probabilistic forecasting of complex dynamical systems. No deep mathematical background required—just an appreciation for probability, AI, and the art of modeling complex systems!