Publication - Trapped in texture bias? A large scale comparison of deep instance segmentation

In the context of the Ph.D. research of Johannes Theodoridis and the Bachelorthesis of Jessica Hofmann, we analysed the robustness of deep learning models for image segmentation. The results have been presented at the ECCV 2022 conference in Tel Aviv and published in the Springer conference proceedings. Here is the abstract: Do deep learning models for instance segmentation generalize to novel objects in a systematic way? For classification, such behavior has been questioned. [Read More]

Publication - Deep feature learning of in-cylinder flow fields to analyze cycle-to-cycle variations in an SI engine

The joint work of Mercedes Benz AG, Technical University Darmstadt and HdM Stuttgart on Deep feature learning of in-cylinder flow fields has been published in the International Journal of Engine Research. The research work was primarily driven by the Masterthesis of our CSM student Daniel Dreher. Here is the abstract: Machine learning (ML) models based on a large data set of in-cylinder flow fields of an IC engine obtained by high-speed particle image velocimetry allow the identification of relevant flow structures underlying cycle-to-cycle variations of engine performance. [Read More]

Publication - Expanding dynamic range in a single-shot image through a sparse grid of low exposure pixels

The joint work of Leon Eisemann (Student CSM), Jan Fröhlich (AM), Axel Hartz (AM) and Johannes Maucher (MI/CSM) on ML-based dynamic range expanding in single-shot images, has been presented recently at IS&T International Symposium on Electronic Imaging 2020 in San Francisco. The paper can be downloaded from here. Here is the abstract: Camera sensors are physically restricted in the amount of luminance which can be captured at once. To achieve a higher dynamic range, multiple exposures are typically combined. [Read More]

Publication - Enhancing Decision Tree based Interpretation of Deep Neural Networks through L1-Orthogonal Regularization

Our former student Nina Schaaf, published the results of her master-thesis at the 18th International Conference on Machine Learning and Applications - ICMLA 2019, Boca Raton, Florida. The thesis has been jointly supervised by Professor Marco Huber and Professor Johannes Maucher. Here is the paper’s abstract: One obstacle that so far prevents the introduction of machine learning models primarily in critical areas is the lack of explainability. In this work, a practicable approach of gaining explainability of deep artificial neural networks (NN) using an interpretable surrogate model based on decision trees is presented. [Read More]

Publication on Low-Resource Text Classification using Domain-Adversarial Learning

Recently, my colleague and Ph.D.-candidate Daniel Griesshaber published our joined work with Professor Ngoc Thang Vu at the 6th Conference on Statistical Language and Speech Processing (SLSP) in Brussels. Here is the paper’s abstract: Deep learning techniques have recently shown to be successful in many natural language processing tasks forming state-of-the-art systems. They require, however, a large amount of annotated data which is often missing. This paper explores the use of domain-adversarial learning as a regularizer to avoid overfitting when training domain invariant features for deep, complex neural network in low-resource and zero-resource settings in new target domains or languages. [Read More]

IEEE Publication Symbolic Reasoning for Hearthstone

In the context of Andreas Stiegler’s Ph.D. we researched Reasoning in Game AI. Andreas particularly focused on reasoning in the card-trading game Hearthstone. Our findings have now been published in the Journal IEEE Transactions on Games - Volume: 10, Issue:2, June 2018. Many thanks to Andreas and the other co-authors Keshav P. Dahal and Daniel Livingstone. Here is the paper’s abstract: Trading-card games are an interesting problem domain for Game AI, as they feature some challenges, such as highly variable game mechanics, that are not encountered in this intensity in many other genres. [Read More]

Publication in European Journal of Applied Physiology

Machine Learning is a quite universal science. Or better: The field of Machine Learning applications is currently growing rapidly. Right now, physiology is not a prime ML application-field, but I believe that there is much potential for intelligent algorithms to provide new insights in many medical disciplines. One proof for this believe is the fact that Machine Learners like me can publish in Physiologial Journals. Yes! This is my first publication in a Journal of Physiology: Recovery of the cardiac autonomic nervous and vascular system after maximal cardiopulmonary exercise testing in recreational athletes. [Read More]