Data Science Research Group
of the
University of Applied Sciences Zwickau
Who we are.
Research Results
The following cards show some of our research results, especially some of our software projects.
Tensor-Calculus (TC)
TC is a scientific computing library for machine learning. TC establishes practical access to neural tensor formats (NTF). NTF have many applications in all fields of machine learning, where machine learning mainly deals with function approximations.
Discrete Optimization Extension (DOpE)
A library for solving Discrete Minimization Problems. Dope provides Methods like AdaptiveCrossSearch, BayesSearch, GridSearch and RandomSearch. It is used for model optimization.
Bias Variance Agent (BiVA)
Biva library serves as a tool to handle bias-variance tradeoff while solving supervised learning problems. Agent provides API-independent approach for model exploration, evaluation and hyperparameter optimisation.
The Team
These are the people working in our research group.
- Data Science
- Development of Methods for Machine Learning and Artificial Intelligence
- Analysis and Numerics
- Numerical Treatment of High Dimensional Problems by Means of Tensor Format Representations
- Mathematics for Engineers and in the Sciences
- since 3/2017
Professor of Mathematics at WHZ - 04/2014-02/2017
Professor of Numerical and Applied Analysis at RWTH Aachen - 10/2013-03/2014
Researcher at TU Berlin, Prof. Dr. Harry Yserentant. - 01/2008-09/2013
Researcher at Max Planck Institute for Mathematics in the Sciences, Leipzig, Prof. Dr. Dr. h.c. Hackbusch. - 04/2005-12/2007
PhD Student at the Max Planck Institute for Mathematics in the Sciences, Leipzig, Supervisor Prof. Dr. Dr. h.c. W. Hackbusch
Christian Walter
Research Assistant
Dr. rer. nat. Dominic Schneider
Research Assistant
Kostiantyn Pysanyi
Research Assistant
Luma AlHajjar
Student Assistant
Mahmoud Hafez
Student Assistant
Marcel Becker
Research Assistant
Michaela Banert
PR Coordinator
Moritz Schwab
Student Assistant
Saad Ahmad
Student Assistant
Tobias Haubold
Research Assistant
Former Members
Antje Holtz
Christian Ahlswede
David Gadsch
Katsiaryna Radschanka
Laura Braun
Samuel Werner
Tommy Hartmann
Research Projects
The following projects are an excerpt of our research work and funding.
KI-Lab-EmCo
The aim of the project is to set up a holistic AI laboratory to use modern methods of condition monitoring by means of machine learning. Current methods as well as new methods using neural tensor formats are taken into account.
MoITra
The aim of the project is to develop a radio-based driverless transport system for intralogistics, which, due to the deliberate logic of tensor formats, can on the one hand be highly fail-safe and on the other hand quickly and adaptively handle new tasks.
Saxony5 - CCL KI
The aim of the project is the implementation of a co-creation lab (CCL) "Artificial Intelligence" in the joint project "Saxony5". This CCL is intended to promote regional exchange on the topics
K-M-I
The K-M-I project is funded by the German Federal Ministry of Education and Research (BMBF). The aim is to research how AI measures can be used to shape work in the region of the Central German coalfield and in West Saxony. This can be implemented, among other things, by supporting intelligent assistance systems in production planning and control or in the maintenance and servicing of complex plants.
KI-StudiUm
The aim of the project is to transfer some methods of AI as a supporting technology into the regular operation of teaching and administration at WHZ. At the same time, the expansion of digital, international and interdisciplinary study programs in breadth and depth as well as the development of AI qualification programs for university employees will take place.
Accident reconstruction in GIDAS by AI
The aim of the project is to work out the current reconstruction process and the potential for improvement of the reconstruction in GIDAS. The reconstruction shall be made more efficient and transparent with regard to tolerances of the results of the pre-crash phase and the kinematic parameters of the collision by using methods of AI.
Publications
This is an excerpt of our publications.
(10) Tensor Approximation and Decomposition
Iterative algorithms for the post-processing of high-dimensional data
PUBLISHED: 06/2020 - Journal of Computational Physics, 410, 109396 AUTHORS: Mike Espig, Wolfgang Hackbusch, Alexander Litvinenko, Hermann G. Matthies, Elmar Zander DOI: 10.1016/j.jcp.2020.109396 LINK: https://www.sciencedirect.com/science/article/pii/S0021999120301704 On the Convergence of Alternating Least Squares Optimisation in Tensor Format Representations
PUBLISHED: 05/2015 - preprint AUTHORS: Mike Espig, Wolfgang Hackbusch, Aram Khachatryan ARXIV: 1506.00062 LINK: https://www.igpm.rwth-aachen.de/forschung/preprints/423 Convergence of Alternating Least Squares Optimisation for Rank-One Approximation to High Order Tensors
PUBLISHED: 11/2014 - preprint AUTHORS: Mike Espig, Aram Khachatryan ARXIV: 1503.05431 LINK: https://www.igpm.rwth-aachen.de/forschung/preprints/412 Efficient low-rank approximation of the stochastic Galerkin matrix in tensor formats
PUBLISHED: 03/2014 - Computers & Mathematics with Applications, 67(4), 818-829 AUTHORS: Mike Espig, Wolfgang Hackbusch, Alexander Litvinenko, Hermann G. Matthies, Philipp Wähnert DOI: 10.1016/j.camwa.2012.10.008 LINK: https://www.mis.mpg.de/de/publications/mis-preprints/2012/2012-28.html Variational Calculus with Sums of Elementary Tensors of Fixed Rank
PUBLISHED: 11/2012 - Numerische Mathematik, 122(3), 469-488 AUTHORS: Mike Espig, Wolfgang Hackbusch, Thorsten Rohwedder, Reinhold Schneider DOI: 10.1007/s00211-012-0464-x LINK: https://www.mis.mpg.de/de/publications/preprints/2009/prepr2009-52.html Efficient Analysis of High Dimensional Data in Tensor Formats
PUBLISHED: 08/2012 - In Sparse Grids and Applications (pp. 31-56). Springer, Berlin, Heidelberg AUTHORS: Mike Espig, Wolfgang Hackbusch, Alexander Litvinenko, Hermann G. Matthies, Elmar Zander DOI: 10.1007/978-3-642-31703-3_2 LINK: https://www.mis.mpg.de/de/publications/preprints/2011/2011-62.html A Regularized Newton method for the Efficient Approximation of Tensors Represented in the Canonical Tensor Format
PUBLISHED: 05/2012 - Numerische Mathematik, 122(3), 489-525 AUTHORS: Mike Espig, Wolfgang Hackbusch DOI: 10.1007/s00211-012-0465-9 LINK: https://www.mis.mpg.de/de/publications/preprints/2009/prepr2009-52.html A note on approximation in Tensor Chain format
PUBLISHED: 04/2012 - Computing and Visualization in Science, 15(6), 331-344 AUTHORS: Mike Espig, Kishore Kumar Naraparaju, Jan Schneider DOI: 10.1007/s00791-014-0218-7 LINK: https://www.mis.mpg.de/de/publications/mis-preprints/2012/2012-16.html Optimization Problems in Contracted Tensor Networks
PUBLISHED: 08/2011 - Computing and visualization in science, 14(6), 271-285 AUTHORS: Mike Espig, Wolfgang Hackbusch, Stefan Handschuh, Reinhold Schneider DOI: 10.1007/s00791-012-0183-y LINK: https://www.mis.mpg.de/de/publications/preprints/2011/2011-66.html Black Box Low Tensor Rank Approximation using Fibre-Crosses
PUBLISHED: 10/2009 - Constructive approximation, 30(3), 557 AUTHORS: Mike Espig, Lars Grasedyck, Wolfgang Hackbusch DOI: 10.1007/s00365-009-9076-9 LINK: https://www.mis.mpg.de/de/publications/mis-preprints/2008/prepr2008-60.html
(6) Quantum Chemistry
Tensor Representation Techniques for Full Configuration Interaction: A Fock Space Approach Using the Canonical Product Format
PUBLISHED: 06/2016 - The Journal of chemical physics, 144(24), 244102 AUTHORS: Karl-Heinz Böhm, Mike Espig, Alexander A. Auer DOI: 10.1063/1.4953665 LINK: https://www.igpm.rwth-aachen.de/forschung/preprints/444 Mesh-free canonical tensor products for six-dimensional density matrix: computation of kinetic energy
PUBLISHED: 06/2016 - Computing and Visualization in Science, 17(6), 267-275 AUTHORS: Sambasiva Rao Chinnamsetty, Mike Espig, Wolfgang Hackbusch DOI: 10.1007/s00791-016-0263-5 LINK: https://www.mis.mpg.de/de/publications/preprints/2013/2013-70.html Tensor Representation Techniques in post-Hartree Fock Methods: Matrix Product State Tensor Format
PUBLISHED: 04/2013 - Molecular physics, 111(16-17), 2398-2413 AUTHORS: Udo Benedikt, Henry Auer, Mike Espig, Wolfgang Hackbusch, Alexander Auer DOI: 10.1080/00268976.2013.798433 LINK: https://www.mis.mpg.de/de/publications/preprints/2013/2013-8.html Tensor Decomposition in post-Hartree Fock Methods
PUBLISHED: 02/2011 - The journal of chemical physics, 134(5), 054118 AUTHORS: Udo Benedikt, Alexander A. Auer, Mike Espig, Wolfgang Hackbusch DOI: 10.1063/1.3514201 LINK: https://www.mis.mpg.de/de/publications/preprints/2013/2013-8.html Canonical tensor products as a generalization of Gaussian-type orbitals
PUBLISHED: 05/2010 - International journal of research in physical chemistry and chemical physics, 224(3-4), 681-694 AUTHORS: Sambasiva Rao Chinnamsetty, Mike Espig, Heinz-Jürgen Flad, Wolfgang Hackbusch DOI: 10.1524/zpch.2010.6131 LINK: https://www.mis.mpg.de/de/publications/preprints/2009/prepr2009-29.html Tensor product approximation with optimal rank in quantum chemistry
PUBLISHED: 08/2007 - The journal of chemical physics, 127(8), 084110 AUTHORS: Sambasiva Rao Chinnamsetty, Mike Espig, Boris N. Khoromskij, Wolfgang Hackbusch DOI: 10.1063/1.2761871 LINK: https://www.mis.mpg.de/de/publications/mis-preprints/2007/prepr2007-41.html
(1) Hierarchical Matrices
On the Robustness of Elliptic Resolvents Computed by means of the Technique of Hierarchical Matric
PUBLISHED: 12/2008 - Applied Numerical Mathematics, 58(12), 1844–1851 AUTHORS: Mike Espig, Wolfgang Hackbusch DOI: 10.1016/j.apnum.2007.11.006 LINK: https://www.mis.mpg.de/de/publications/mis-preprints/2007/prepr2007-11.html
Our Partners
Here are our partners who support us in our work.
