WATCH VIDEONumerical methods and high-performance computing have rapidly advanced in the past few decades, enabling accurate simulations of complex thermo-fluid systems in many engineering problems. However, faster and higher resolution simulations, particularly for design, optimization, and online control, are of increasing demand, requiring novel approaches to simulating multi-scale, multi-physics, thermo-fluid systems. In the past few years, a number of studies have explored how deep learning might accelerate computational fluid dynamics (CFD). In this talk, I will discuss some of the recent advances in this area, and in particular will present our group’s work on 1) using recurrent neural networks for data-driven integration of some of the computationally demanding equations in a CFD solver, and 2) using transfer learning to improve generalization (i.e., extrapolation) of a deep learning-based CFD solver. Examples from a multi-scale Lorenz system and plans for future applications to pipe flows and Rayleigh-Benard convection will be discussed.