Machine Learning at the Flatiron Institute Seminar: Dmitri Kochkov

Title: Numerical methods + ML for simulation of turbulent systems

Abstract: Simulations of turbulent flows are routinely used in fundamental research and have numerous practical applications such as climate modeling, weather forecasting and industrial engineering. While numerical analysis provides a broad and systematic spectrum of techniques for simulating such systems, there has been a lot of interest in applying machine learning to address the limitations of traditional approaches. In this talk I’ll discuss the class of hybrid modeling approaches that attempt to leverage strengths of both numerical and machine learning methods. In particular, I’ll summarize a few published results from our recent efforts on simulation of fluid flows in 2 dimensions and share my perspective on applications of such modeling techniques in the context of atmospheric modeling.