The first set of RDs is in the physical sciences, where use of scientific computations is widespread and machine learning approaches are being progressively explored. A ubiquitous problem in physical sciences is parameter identification of physically based models. To account for data uncertainty and identify probability distribution functions of parameter values that are consistent with available data, statistical regression, particularly Bayesian regression, is an appropriate framework for analysis. As described below, uncertainty quantification is already in progress in many areas and now the aim is shifting to efficiently propagating the identified uncertainties in forward models.
- Learning geospatial data with Gaussian Random Field models
- Gravity field estimation from satellite data
- Uncertainty, precision and reliability of eco-hydrological models
- Uncertainty propagation for viscoelastic mechanics models
- Improving powder deposition in additive manufacturing through machine learning
- Data-driven identification of governing equations in continuum mechanics
In physics, rational design of molecular crystals requires precise control and understanding of the relationship between composition, structure and properties, which is possible through clustering and dimensional reduction of simulation data. There is a need to develop the big data and HPC tools required for in silico discovery of molecular materials.
- Big Data Platform for Advanced Molecular Crystals Simulations and Analytics
New approaches in Applied Mathematics and Statistics combined with revolutionary concepts in Data Science open up opportunities for truly scaling computational algorithms. Such advanced techniques, combined with Model Order Reduction and data-oriented advanced discretisation methods, allow to break through the barriers of existing predictive models and associated simulation paths.
- MapReduce for reduced order modelling of multiphysics problems
- Open-Source Platform for Free Boundary Problems