Gabriel Weymouth | Technical University Delft

In the last few years machine learning has made amazing strides in virtually instantaneous predictions for language modelling and image generation. Many scientific and engineering disciplines, such as fluid dynamics, are very computationally expensive to simulate, leading to the natural goal of using machine learning tools to accelerate these predictions. But can tools developed for qualitative predictions with millions of examples be used effectively on quantitative engineering problems with sparse data? Can physics be incorporated into large ML predictions without slowing down the model or hindering its generalization? This seminar covers some of Gabriel's efforts towards machine-learning-accelerated flow solvers, turbulent flow predictions, and maneuvering optimization.

This talk was presented during a hybrid event at the TU Delft. You can find a recording of this past event at Vimeo.