This user has not added any information to their profile yet.
You can reach me via e-mail at:
bgeiger<AT>know-center.at OR geiger<AT>ieee.org
I studied Electrical Engineering with focus on Communications between 2004 and 2009 at Graz University of Technology. I then joined the Signal Processing and Speech Communication Lab (SPSC) as a Research & Teaching Associate and worked on my PhD thesis in the intersection between signal processing and information theory. In November 2014, I joined the Institute of Communications Engineering of the Technical University of Munich. Since 2018 I am a Senior Researcher at the Know-Center GmbH, and I lead its Machine Learning Group since April 2020. I lecturer at the Graz University of Technology and at the University of Graz.
In my leisure time I enjoy reading a good book. My other hobbies are running, Geocaching, and playing Go (~12 kyu; you can challenge me - sliver1984 - at DGS).
My research focus is on theory-inspired machine learning, information-theoretic approaches to the design and analysis of machine learning systems (especially deep learning), and model reduction for Markov models. For my most recent publications, see below or check out my Google Scholar profile and my preprints on arXiv.
I am extremely lucky and grateful for being able to do research together with
- Joao de Freitas (PhD Candidate)
- Johannes Hoffer (PhD Candidate, external)
- Andreas Ofner (PhD Candidate)
- Franz Rohrhofer (PhD Candidate)
- Maximilian Toller (Senior Researcher)
Open Master Theses and Student Projects
- Is MC Dropout Sensitive to Dead Neurons?
- Information Plane Analysis of Binary/Quantized Neural Networks
- Meaning of Hypothesis tests based on finite populations
- On the Information Loss in Non-Injective Normalizing Flows
- Causal Training Mechanism in Physics-Informed Neural Networks for Dynamical Systems
- Information Plane Analysis of Variational Information Bottleneck
- Does Variational Information Bottleneck lead to Sparse Weight Matrices?
- 2023-05-15: Quite some things have happened recently: One paper on robust Bayesian optimization at Computers and Industrial Engineering (congrats, Johannes!), one on loss weighting schemes for learning chemical reactions in neural networks that won the Best Paper Award at Scientific Computing (congrats, Franz!), and one paper on information plane analyses at ICLR -- Reasons to celebrate!
- 2023-01-23: Franz' paper on how fixed points of dynamical systems affect PINN training has been accepted in the Transactions on Machine Learning Research! Congrats, Franz!
- 2022-12-15: Our former colleague and external PhD candidate Johannes has recently published a paper on the Bayesian optimization of process chains in CIRP Journal of Manufacturing Science and Technology. Congrats, Johannes!
- 2022-12-01: Our group has published two papers in the prestigious IEEE Transactions on Neural Networks and Learning Systems, dealing with information-theoretic analyses of neural networks. The papers focus on information plane analysis and individual neuron importance, respectively. The latter work is a collaboration with Rana Ali Amjad (Amazon) and Kairen Liu (Rhode & Schwarz).
- 2022-10-03: Joao's paper on hierarchical representations for multi-task learning, which was presented at IJCNN in Padua in July, is now available on IEEE Xplore -- congrats, Joao! If you want to read it without paywall, you can find a copy on arXiv.
- 2022-08-12: Andy's paper on reconstructing in-cylinder pressure from knock sensors is published at Elsevier's Mechanical Systems and Signal Processing -- congrats, Andy!
- 2022-06-28: Our colleague Sophie has successfully completed her Master thesis with the title "Physics Informed Neural Networks: analysis for a dynamical system - the double pendulum". Excellent work, congratulations!
- 2022-05-10: Our colleague Sophie has just published a paper on semi-supervised clustering using Markov aggregation at the ACM SAC and on arXiv. This is joint work with Marek Smieja from Jagiellonian University of Krakow. Congrats, Sophie, and thanks for the amazing video!
- 2022-05-06: Another publication: We just published a paper on synthetic graph generation at the ACM Web Conf and on arXiv. Congrats to all authors!
- 2022-04-26: Our groups tutorial/survey paper on theory-inspired machine learning is now published in Welding in the World.
- 2022-02-25: Our colleague Max has successfully defended his PhD thesis. Congratulations, Dr. Toller!
- 2022-01-31: Another publication from our DiSpecs project: Our executable (!) paper on sentiment analysis in digital literary studies has been published by Melusina Press.
- 2022-01-21: Johannes' paper on Gaussian surrogate models for structural mechanics has been published at Applied Sciences -- congrats, Johannes!
- 2022-01-18: Andy's paper on the detection of engine knock using theory-inspired convolutional neural networks has been accepted for publication in the IEEE/ASME Transactions on Mechatronics -- congrats, Andy! A preprint of this joint work with Achilles Kefalas (IVT, TU Graz) and Stefan Posch (LEC GmbH) is available at arXiv.
- 2022-01-10: Synwalk, an information-theoretic approach to community detection is now published in Data Mining and Knowledge Discovery and will be presented at ECML-PKDD 2022. This joint work with Christian Toth (SPSC, TU Graz) and Denis Helic (ISDS, TU Graz) grew out of Christian's Master's thesis -- congrats!
- 2021-12-15: Our Master student and colleague Sophie Steger got her paper on semi-supervised clustering via information-theoretic Markov aggregation accepted at ACM SAC (see extended version on arXiv) -- congrats, Sophie! Joint work with Marek Śmieja from the Jagiellonian University of Krakow.
- 2021-09-03: SeGMA, our semi-supervised auto-encoder is now published in the IEEE Trans. Neural Networks and Learning Systems. Joint work with Marek Śmieja, Maciej Wołczyk, and Jacek Tabor from the Jagiellonian University of Krakow.
- 2021-07-09: I was honored to deliver the Portevin Lecture at the IIW International Conference on Artificial Intelligence to innovate Welding and Joining, where I presented work on theory-inspired ML jointly done with Andreas, Franz, and Johannes.
- 2021-07: Cluster Purging: Our rate-distortion theory-based approach to outlier detection was accepted at IEEE Trans. Knowledge and Data Engineering. Congrats, Max!