Dr. Stefanie Günther

Chair for Scientific Computing
TU Kaiserslautern Bldg/Geb 34, Paul-Ehrlich-Straße
67663 Kaiserslautern, Germany

Office:  36-407
Phone: +49 (0)631 205 3428
Fax:   +49 (0)631 205 3056
Email:   stefanie.guenther@scicomp.uni-kl.de

s-img_2599

Research Interests


I am focusing on the simultaneous methods for solving optimization problems with unsteady PDE constraints – mainly for aerodynamics application. One major difficulty arises from sensitivity computation with turbulent, i.e. chaotic dynamics. Here, a regularization technique is required in order to compute meaningful sensitivities – the so-called Least Squares Shadowing problem.

I am also interested in parallel-in-time integration methods for solving unsteady PDEs and their integration into a simultaneous optimization framework. I have developed an adjoint solver for the multigrid-in-time software library XBraid, which distributes workload onto multiple processors along the time domain. Its usage in a simultaneous One-shot optimization framework has great potential for speeding up the optimization process with unsteady dynamics. Currently. I’m developing an interface for the open source CFD suite SU2 which provides a great test base for the parallel-in-time integration and optimization approach.

Recently, I got engaged in a very exciting new topic: Simultaneous parallel-in-layer training for Deep Neural Networks. Motivated by the interpretation of deep residual network training as an optimal control problem, I’m exploring the potential of multigrid-in-layer methods withing a simultaneous training process.

In short:

  • Simultaneous optimization with unsteady PDs
  • Sensitivity computation for chaotic dynamics
  • Parallel-in-time integration and optimization
  • Simultaneous Parallel-in-Layer training for deep residual networks

Find me on github: steffi7574

Resume


  • Since 2017: Postdoc at Chair for Scientific Computing, TU Kaiserslautern
  • 2017: PhD in Applied Mathematics with N. Gauger, RWTH Aachen University
  • 2014 – 2017: Research assistant at Chair for Scientific Computing, TU Kaiserslautern
  • 2012 – 2014: Research assistant at MathCCES, RWTH Aachen University
  • 2012: Diploma (equiv. to M.Sc.) in Applied Mathematics with V. Schulz, University of Trier
  • 2006 – 2012: Studies in Applied Mathematics with minor in Computer Science, University of Trier

Publications


PhD Thesis: Simultaneous Optimization with Unsteady Partial Differential Equations, RWTH Aachen University, 2017
Diploma Thesis: Simultane Optimierung unter PDE-Nebenbedingungen – Ein Vergleich zweier One-Shot-Methoden, University of Trier, 2012

2019

S. Günther, L. Ruthotto, J.B. Schroder, E.C. Cyr, N.R. Gauger (2019): Layer-Parallel Training of Deep Residual Neural Networks. In: (submitted to arXiv v1 on 11.12.2018) submitted to SIAM Journal on Mathematics of Data Science (SIMODS), 2019. (Type: Article | Links | BibTeX)

2018

S. Günther, N.R. Gauger, J.B. Schroder (2018): A Non-Intrusive Parallel-in-Time Adjoint Solver with the XBraid Library. In: Computing and Visualization in Science, 19 (3), pp. 85-95, 2018, ISSN: 1433-0369. (Type: Article | Links | BibTeX)
S. Günther, N. R. Gauger, J. B. Schroder (2018): A Non-Intrusive Parallel-in-Time Approach for Simultaneous Optimization with Unsteady PDEs. In: Optimization Methods and Software, to appear, 2018. (Type: Article | Links | BibTeX)

2017

S. Günther, N.R. Gauger, Q. Wang (2017): A framework for simultaneous aerodynamic design optimization in the presence of chaos. In: Journal of Computational Physics, 328 pp. 387-398, 2017, (DOI 10.1016/j.jcp.2016.10.043). (Type: Article | Links | BibTeX)

2016

S. Günther, N.R. Gauger, Q. Wang (2016): Simultaneous One-shot optimization with unsteady PDEs. In: Journal of Computational and Applied Mathematics, (294), pp. 12-22, 2016, (arXiv preprint http://arxiv.org/pdf/1503.00884v1.pdf). (Type: Article | Links | BibTeX)

2015

S. Günther, N.R. Gauger, Q. Wang (2015): Extension of the One-Shot Method for Optimal Control with Unsteady PDEs. In: Computational Methods in Applied Sciences, 36 pp. 127-142, 2015. (Type: Article | BibTeX)

2014

T. Bosse, N.R. Gauger, A. Griewank, S. Günther, V. Schulz (2014): One-Shot Approaches to Design Optimzation. In: International Series of Numerical Mathematics, 165 pp. 43-66, 2014, (DOI:10. 1007/978-3-319-05083-6_5). (Type: Article | BibTeX)
T. Bosse, N.R. Gauger, A. Griewank, S. Günther, L. Kaland, et al. (2014): Optimal Design with Bounded Retardation for Problems with Non-separable Adjoints. In: International Series of Numerical Mathematics, 165 pp. 67–84, 2014, (DOI:10. 1007/978-3-319-05083-6_6). (Type: Article | BibTeX)

Talks

  • S. Günther, L. Ruthotto, J.B. Schroder, E.Cyr, N.R. Gauger: Simultaneous Layer-Parallel Training for Deep Residual Networks, SIAM CSE 2019, Feb 2019, Spokane, WA, USA
  • S. Günther, L. Ruthotto, J.B. Schroder, E.Cyr, N.R. Gauger: Simultaneous Layer-Parallel Training for Deep Residual Networks, Fraunhofer ITWM, Deep Learning Seminar, Feb 2019, Kaiserslautern, Germany
  • S. Günther, L. Ruthotto, J.B. Schroder, E.Cyr, N.R. Gauger: Simultaneous Parallel-in-Layer Training of Deep Residual Networks, KU Leuven, Seminar at the NUMA (Numerical Analysis and Applied Mathematics) section, Nov 2018, Leuven, Belgium
  • S. Günther, L. Ruthotto, J.B. Schroder, N.R. Gauger: Simultaneous Parallel-in-Layer Optimization for Training of Deep Residual Networks, EUCCO 2018, Sep 2018, Trier, Germany
  • S. Günther, N.R. Gauger, J.B. Schroder: A Non-Intrusive Parallel-in-Time Method for Simultaneous Optimization with Unsteady PDEs, IFIP 2018, July 2018, Essen, Germany
  • S. Günther, N.R. Gauger, J.B. Schroder: Time-Parallel Simultaneous One-shot Optimization, 6th Workshop on Parallel-in-Time Integration Methods, Okt 2017, Monte Verita, Swizerland
  • S. Günther, N.R. Gauger, J.B. Schroder, Q. Wang: Simultaneous Parallel-in-Time Optimization with Unsteady PDEs, SIAM CSE Conference, Feb 2017, Atlanta, USA
  • S. Günther, N.R. Gauger, J.B. Schroder: Adjoint Sensitivity Computation for the Parallel Multigrid Reduction in Time Software Library XBraid, 5th Workshop on Parallel-in-Time Integration, Nov 2016, Banff, Canada
  • S. Günther, N.R. Gauger, Q. Wang: Simultaneous PDE-constrained Optimization in Chaotic Dynamics – A framework for optimal control of highly separated flows, GAMM Annual Meeting, March 2016, Braunschweig, Germany
  • S. Günther, N.R. Gauger, Q. Wang: Fixed-Point Iterations for Simultaneous One-Shot Optimization of Unsteady Flows, SIAM CSE 2015, March 14-18, Salt Lake City, USA
  • S. Günther, N.R. Gauger, Q. Wang: Towards Efficient One-Shot Optimization with Unsteady Navier-Stokes Equations, GAMM 2014, March 10-14 2014, Erlangen, Germany
  • S. Günther, N.R. Gauger, Q. Wang: Extension of the One-Shot Method for Optimal Control with Unsteady PDEs, EUROGEN 2013, October 7-10 2013, Las Palmas de Gran Canaria, Spain
  • S.Günther, N.R. Gauger: One-Shot Optimization with unsteady PDE constraints, Massachusetts Institute of Technology, Department of Aeronautics and Astronautics, March 04 2014, Cambridge, USA
  • S. Günther, E. Özkaya, A. Nemili, N.R. Gauger: Optimal Active Separation Control on Airfoils using Discrete Adjoint Approach, SIAM CSE 2013, February 25 – March 1 2013, Boston, USA