Deep Learning for Optimal Reduced-Order Scientific Computation  

Project description

Scientific Machine Learning is becoming a rapidly growing discipline within the Computational and Data Sciences. It combines Scientific Computation, which focussed on the numerical simulation of mathematical models from the applied sciences, and Machine Learning, which focusses on algorithms for mathematical models that are data driven. While large-scale numerical simulations are extremely important for predictions of real-world problems, they can be computationally excessive, requiring tremendous computing power. This is where machine-learning algorithms can provide a much-needed solution: Reduced-order models can be learned that allow huge savings in computational costs while remaining accurate in relevant quantities of interest. In this project, the student will explore the above Scientific Machine Learning paradigm by utilising the power of Deep Neural Networks to systematically construct optimal reduced-order models for Partial Differential Equations.