Unfortunately, we have reached the maximum number of participants for this event and are no longer accepting new registrations
Session 1: Neural ODEs
Dr. Marvin Höge
Between pure machine-learning and pure
mechanistic models, hybrid models offer a middle-ground, combining features
like flexibility of data-driven methods and interpretability of physics-based
relations. In my session, I will introduce Neural ODEs as a powerful member of
this category of models and I will demonstrate and discuss its benefits in
hydrologic modelling and related fields.
Further, I will provide an overview of the
programming language Julia and the package ecosystem that I used for
implementing Neural ODEs and that enables various new routes of scientific
programming.
Session 2: Stochastic
Modelling of Spatial Data
Jun.-Prof. Dr. rer. nat. Marco Oesting
The
course will consist of two parts on "Classical Gaussian Modelling"
and "Modelling of Extreme Events", respectively. In both parts, we will first discuss stochastic modelling of a (typically
environmental) variable at single site, then cover strategies for modelling in space (i.e., the inclusion of spatial dependence) and finally present methods supporting the application of the models to practical problems. These methods range from simulation of the models to
spatial interpolation and prediction.
Session 3: Uncertainty quantification: From basics to high-performance computing
Dr. Linus Seelinger & Dr. Lorenzo Tamellini
Uncertainty quantification (UQ) determines the effect of uncertain data on model predictions or inference. This is crucial in medical decision making, nuclear waste disposal, design of safe aircraft, etc. This session introduces UQ and some state-of-the-art methods, namely sparse grids surrogate modeling and the Sparse Grids Matlab Kit. Moreover, through UM-Bridge, a universal UQ software interface, the exercises cover both basic examples and high-performance applications on a cluster.
created with
Website Builder Software .
