Tobias Windisch
Most of my research is about creating machine learning
systems that further automatize industrial processes
using data from optical sensors. Particularly, I like to
develop mechanical processes build around a machine
learning model.
Projects
Causal representation learning
Causal relation mining on high-dimensional manufacturing data, including images and time-series data. Here, deep data representations are combined with graphical models to unveil relations between manufacturing processes.
Line optimization with reinforcement learning
Research project LineFlow funded by the Bavarian state ministry of research. The goal is to develop reinforcement learning algorithms that automatically optimize assembly lines by adapting process parameters in order to avoid bottlenecks and maximize throughput. Learn more here.Working Group
Kai Müller, M.Sc.
Reinforcement learning for active line control
Reinforcement learning for active line control
Matthias Burkhardt, M.Sc.
Reinforcement learning for active alignment
Reinforcement learning for active alignment
Edgar Wolf, M.Sc.
Drift detection in high-dimensional sensory data
Drift detection in high-dimensional sensory data
Fabian Hueber, B.Sc.
Latent drift detection with Autoencoder
Latent drift detection with Autoencoder
Kilian Führer, B.Sc.
Representation learning for process curves
Representation learning for process curves
Martin Wenzel, B.Sc.
Reinforcement learning for active line control
Reinforcement learning for active line control
Cindy Buhl, M.Sc.
Machine learning for production data
Machine learning for production data
Tobias Schmähling, M.Sc.
Reinforcement learning for active alignment
Reinforcement learning for active alignment
Industrial research partners
Publications
- A method to benchmark high-dimensional process drift detection
with E. Wolf, submitted, arXiv:2409.03669, Code - causalAssembly: Generating Realistic Production Data for Benchmarking Causal Discovery
with K. Göbler, T. Pychynski, S. Sonntag, M. Roth, and M. Drton, Proceedings of Machine Learning Research, 3rd Conference on Causal Learning and Reasoning, Spotlight paper, arXiv:2306.10816, Code - The fiber dimension of a graph
Discrete Mathematics, arXiv:1601.04527 - Heat-bath random walks with Markov bases
with C. Stanley, Advances in Applied Mathematics, arXiv:1605.08386 - Fiber graphs
PhD thesis, OvGU Magdeburg, 01/2017 - Rapid mixing and Markov bases
SIAM Journal on Discrete Mathematics, arXiv:1505.03018 - Parity binomial edge ideals
with T. Kahle and C. Sarmiento, Journal of Algebraic Combinatorics, arXiv:1503.00584 - On the Connectivity of Fiber Graphs
with R. Hemmecke, Journal of Algebraic Statistics, arXiv:1405.0812
Patent applications
- Method and device for aligning a lens system with machine learning
with Benno Geißelmann and Steffen Löwendorf, Robert Bosch GmbH (applicant), November 9, 2021, DE102021212601.A1, US20230147112.A1, CN116107086A, JP2023070666A, KR20230067565A