Show filters Hide filters

Refine your search

Publication Year
1-9 out of 9 results
Change view
  • Sort by:
47:23 Weierstraß-Institut für Angewandte Analysis und Stochastik (WIAS), Technische Informationsbibliothek (TIB) German 2017

"Extreme Fluids" - Some Examples, Challenges and Simulation Techniques for Flow Problems with Complex Rheology

In this talk we discuss numerical simulation techniques for incompressible fluids with complex rheology which means that local flow characteristics may differ significantly by several orders of magnitude, for instance due to non-isothermal behavior and pressure, resp., shear dependent viscosity. Such fluids usually include viscoplastic as well as viscoelastic effects which is typical for yield-stress fluids, granular material as well as polymer melts and kautschuk. Corresponding applications are relevant for polymer processing, but include also viscoplastic lubrication, fracking and macro encapsulation. In this talk, we present special discretization and solver techniques in which case the coupling between the velocity, pressure and additional variables for the stresses, which leads to restrictions for the choice of the FEM approximation spaces, and the (often) hyperbolic nature of the problem are handled with special Finite Element techniques including stabilization methods. The resulting linearized systems inside of outer Newton-like solvers are (special) nonsymmetric saddle point problems which are solved via geometrical multigrid approaches. We illustrate and analyze numerically the presented methodology for well-known benchmark configurations as well as protoypical industrial applications for several nonlinear flow models.
  • Published: 2017
  • Publisher: Weierstraß-Institut für Angewandte Analysis und Stochastik (WIAS), Technische Informationsbibliothek (TIB)
  • Language: German
52:42 Weierstraß-Institut für Angewandte Analysis und Stochastik (WIAS), Technische Informationsbibliothek (TIB) English 2017

Machine learning and applications

Since a few years Machine Learning (ML) has broadened the modeling toolbox for the sciences and industry. The talk will first remind the audience of the main ingredients for applying machine learning. Then various ML applications in the sciences namely Brain Computer Interfaces and Quantum Chemistry will be discussed.
  • Published: 2017
  • Publisher: Weierstraß-Institut für Angewandte Analysis und Stochastik (WIAS), Technische Informationsbibliothek (TIB)
  • Language: English
25:03 Weierstraß-Institut für Angewandte Analysis und Stochastik (WIAS), Technische Informationsbibliothek (TIB) German 2017

Mathematical knowledge management as a route to sustainability in mathematical modeling and simulation

Mathematical modeling and simulation (MMS) has now been established as an essential part of the scientific work in many disciplines. It is common to categorize the involved numerical data and to some extend the corresponding scientific software as research data. Both have their origin in mathematical models. A holistic approach to research data in MMS should cover all three aspects: models, software, and data. Yet it is unclear, whether a suitable management of the mathematical knowledge related to models is possible and how it would look like. In this talk, we outline an approach to address this problem based on a flexiformal representation of the mathematical knowledge in scientific publications and discuss how this can contribute to sustainable research in MMS.
  • Published: 2017
  • Publisher: Weierstraß-Institut für Angewandte Analysis und Stochastik (WIAS), Technische Informationsbibliothek (TIB)
  • Language: German
42:16 Weierstraß-Institut für Angewandte Analysis und Stochastik (WIAS), Technische Informationsbibliothek (TIB) English 2017

An introduction to inverse problems with applications in machine learning

The presentation starts with some motivating examples of Inverse Problems before introducing the general setting. We shortly review the most common regularization approaches (Tikhonov, iteration methods) and sketch some recent developments in sparsity and machine learning. Sparsity refers to additional expert information on the desired reconstruction, namely, that is has a finite expension in some predefined basis or frame. In machine learning we focus on 'multi colored' inverse problems, where part of the application can be formulated by a strict analytical framework but some part of the problem needs to modeled by a data driven approach. Those combined problems can be created by data- driven linear low rank approximations or more general black box models. In particular we review deep learning approaches to inverse problems. Finally, machine learning techniques by themselves are often inverse problems. We highlight basis learning techniques and applications to hyperspectral image analysis.
  • Published: 2017
  • Publisher: Weierstraß-Institut für Angewandte Analysis und Stochastik (WIAS), Technische Informationsbibliothek (TIB)
  • Language: English
30:32 Weierstraß-Institut für Angewandte Analysis und Stochastik (WIAS), Technische Informationsbibliothek (TIB) German 2017

Modeling decision tree training problems as a Mixed Integer Program (MIP) yields optimal decision trees

The problem of constructing an optimal binary decision tree is known to be NP-complete. Therefore most current implementations base on heuristics where local optimality criteria are used. Our approach optimizes decision trees globally by construction a suitable MIP. In recent years both very high computing power and very efficient branch-and-cut algorithms for solving MIPs make the running time more and more realistic for practical applications. We applied our method for the discrimination of tumor samples into distinct telomere maintenance mechanisms (TMM). Telomeres are at the end of the chromosomes and shorten after each replication serving as a crucial check point for protecting cells from unbound replication. Tumor cells circumvent this by either re-evoking the enzyme for elongation or redirecting DNA-repair mechanisms know as alternative TMM. Our approach allowed classifying the tumor samples with an accuracy of 0.95 when using the experimental gold standard (C-circle assays).
  • Published: 2017
  • Publisher: Weierstraß-Institut für Angewandte Analysis und Stochastik (WIAS), Technische Informationsbibliothek (TIB)
  • Language: German
14:53 Weierstraß-Institut für Angewandte Analysis und Stochastik (WIAS), Technische Informationsbibliothek (TIB) English 2017

Estimation of cardiovascular system parameter from real data

This talks shows preliminary results for the estimation of some parameters regarding the cardiovascular system (e.g., vessel mechanical properties, vasculature resistance) using real patient data. The mathematical model is based on efficient one-dimensional network for the blood flow simulation and on the unscented Kalman filtering for the parameter estimation.
  • Published: 2017
  • Publisher: Weierstraß-Institut für Angewandte Analysis und Stochastik (WIAS), Technische Informationsbibliothek (TIB)
  • Language: English
25:37 Weierstraß-Institut für Angewandte Analysis und Stochastik (WIAS), Technische Informationsbibliothek (TIB) German 2017

Thermodynamically consistent modeling of fluids

The motion of fluids is restricted by the 2nd Law of Thermodynamics in multiple manner. This lecture uses historical and contemporary issues to illustrate both the general structure and special properties of thermodynamically consistent modeling.
  • Published: 2017
  • Publisher: Weierstraß-Institut für Angewandte Analysis und Stochastik (WIAS), Technische Informationsbibliothek (TIB)
  • Language: German
16:44 Weierstraß-Institut für Angewandte Analysis und Stochastik (WIAS), Technische Informationsbibliothek (TIB) German 2017

Combining linear Support Vector Machines by constraining them to use the same set of features improves consistency in biomarker discovery for blood infections

Blood infection is highly prevalent in critical ill patients and can lead to sepsis and often death. It can be caused by bacteria or fungi and for appropriate treatment it is mandatory to identify the type of infection early. To find discriminating biomarkers, in situ high throughput gene expression profiling of immune cells after fungal or bacterial infection have been performed. However, these studies showed very heterogeneous results. To find a generic gene signature with discriminative power across all datasets, we implemented linear SVMs basing on Mixed Integer Linear Programming. We combined classifiers constraining them to use the same set of features. Learning with one pair of datasets and applying to the rest of the datasets showed 43?mprovement in consistency of the selected features (genes) while non-decreased classification performance (accuracy: 0.96). The final biomarkers comprised of 19 genes mostly involved in ERK-MAPK signalling being central in immune response.
  • Published: 2017
  • Publisher: Weierstraß-Institut für Angewandte Analysis und Stochastik (WIAS), Technische Informationsbibliothek (TIB)
  • Language: German
14:53 Weierstraß-Institut für Angewandte Analysis und Stochastik (WIAS), Technische Informationsbibliothek (TIB) German 2017

Investigation of phenomena in the Western Baltic Sea

To understand the processes of local phenomena over the Baltic Sea such as Coastal Upwelling or Salinity Inversion, we are coupling an atmosphere and ocean model with the Earth System Modelling Framework (ESMF). For the atmospheric part the operational model of the German Weather Service (ICON) is utilized in a nested limited area mode. The General Estuarine Turbulence Model (GETM) has been chosen for the local ocean model. Typical coupling issues are the different grid schemes of the models and hence, a set and choice of suitable interpolation/regridding methods is required. Within our framework, the state variables (e.g. temperature) and flux data (e.g. heat flux) has to be interpolated from the unstructured triangular grid of ICON to the structured rectangular latitude longitude grid of GETM and vice versa. Furthermore, due to different grids, the land sea masking of each model has to be considered for the interpolation. Additionally, when using a parallel infrastructure, the number of processes has to be chosen such that the coupled model runs well balanced. Since we are using the concurrent structure ESMF is providing, the focus is on the reduce of possible waiting time for each model. The presentation shall give an overview about these issues, how we are addressing them within our coupled model framework and some results of first runs.
  • Published: 2017
  • Publisher: Weierstraß-Institut für Angewandte Analysis und Stochastik (WIAS), Technische Informationsbibliothek (TIB)
  • Language: German
out of 1 pages
Loading...
Feedback

Timings

   63 ms - page object
   51 ms - search
    3 ms - highlighting
    0 ms - highlighting/21902
    2 ms - highlighting/21908
    1 ms - highlighting/30617
    1 ms - highlighting/21916
    1 ms - highlighting/21909
    2 ms - highlighting/21906
    2 ms - highlighting/21899
    1 ms - highlighting/30616
    3 ms - highlighting/21903

Version

AV-Portal 3.8.0 (dec2fe8b0ce2e718d55d6f23ab68f0b2424a1f3f)