Physics-based Machine Learning, Data Science

Hybridized data-integrated and process-based simulation workflows often imply computationally infeasible high-throughput tasks, such as uncertainty quantification, optimization, or Bayesian inference. Physics-based machine learning, such as PINNs or Gaussian Process Emulation, trains a data-driven model based on simulation results and thereby offers a pathway to fast and efficient model evaluation at low error rates. Research on physics-based machine learning includes aspects of both simulation, and data science and will be one of the central research topics addressed in the House of Data-Driven Sciences (HDDS).