Scientific Machine Learning (SciML) is a recent emerging area in scientific computing that investigates the integration of physical modeling and simulation with their data-driven counterpart, aiming to combine the strengths and compensate for the weaknesses of both. In particular, SciML (i) uses numerical models for the improvement of established machine learning techniques, (ii) uses data-driven machine learning techniques to assist numerical simulations, and (iii) creates new high-performance computational methodologies by employing aspects from both sides.
The MIA team in SciML has specialized interests in Bayesian learning for dynamical systems, physically consistent model reduction, and physics-informed deep learning. In particular, research efforts are being made to synergize deep learning and Gaussian processes to enable robust data-driven discovery of low-dimensional dynamics from various measurement data, while certifying computational reliability with an endowment of uncertainty quantification.
People