On-the-fly Model Modification, Error Control, and Simulation Adaptivity

Project Network 5

RQs linking accuracy, precision, and resources

We aim to achieve large-scale simulations of unprecedented reliability and efficiency in the triangle spanned by the accuracy demand for small systematic errors, the precision challenge to attain low stochastic errors, and the resource limitations imposed by computing power and available data. Classical scientific computing (accuracy vs. resources), uncertainty quantification (precision vs. resources) and stochastic modeling (precision vs. accuracy) can be associated with the edges of the triangle. Even along single edges, it is unknown how to determine optimal trade-offs in view of multiple optimization criteria. In the expectation of a more powerful yet also more heterogeneous hardware structure, we plan on tailoring and dynamically optimizing models and simulations in this triangle. The key issue will be the integration of information derived not only from experimental data but also from metadata, i.e., data about the simulations themselves.

Research Questions

RQ 1 Accuracy vs. resources: How can we dynamically adapt towards specific accuracy targets, given the hardware, time and power constraints imposed by a heterogeneous, yet undefined, hardware landscape?

RQ 2 Precision vs. resources: How can we quantify uncertainties and dynamically ensure confidence in simulations given randomness in observational data, limitations of computational resources and incomplete models?

RQ 3 Precision vs. accuracy: How can we construct and dynamically adjust stochastic models with optimized complexity despite limited domain knowledge and limited availability of calibration data?

RQ 4 The triangle: How can we ultimately quantify and adaptively approach global minima for resource requirements given a prescribed total error (accuracy plus precision) target?

Project Overview

PN 5-1 Optimal Multidimensional Accuracy-Resource Tradeoff
PN 5-2 Data-driven optimisation algorithms for local dynamic model adaptivity
PN 5-3 Efficiency Improvements by Code Intrinsic Uncertainty Quantification via Monte Carlo Methods
PN 5-4 Processing uncertain microstructural data
PN 5-6 Physics-informed ANNs for dynamic, distributed and stochastic systems
PN 5-6 (II) Bayesian, Causal, Universal Differential Equation Learner
PN 5-7 Uncertainty Quantification by Physics- and Data-Based Models for Mechanical Systems
PN 5-8 Exploiting structural knowledge in (nonlinear) PDE problems for efficient deployment of the learning abilities of ANNs
PN 5-8 (II) Effective uncertainty quantification and ANN dynamics via amplitude equations
PN 5-9 Adaptive and Flexible Macro-Micro Coupling Software
PN 5-10 Bayesian Multiscale Spatio-Temporal Modelling of Extreme Events
PN 5-11 Uncertainty quantification for data-limited inverse problems via efficient probabilistic ML methods
PN 5-12 Machine Physics Learning for spatio-temporal systems: ANN generalization using adaptive grids, domain decomposition and parallelization

Associated Projects

Project Network Coordinators

This image shows Andrea Barth

Andrea Barth

Prof. Dr.

[Photo: SimTech/Max Kovalenko]

This image shows Dominik Göddeke

Dominik Göddeke

Prof. Dr. rer. nat.

[Photo: SimTech/Max Kovalenko]

To the top of the page