ML2R became the Lamarr Institute – find all current information here!
Introducing the Scientists
Get an overview of the scientific ML2R team! The scientists introduce themselves with a short profile and reveal topics they are currently working on, which Machine Learning aspects they are particularly interested in and to which of the ML2R research focus areas their work contributes.
Dr. Maram Akila
Research Focus:
Trustworthy Machine Learning
What problems are you currently working on?
- Differentiation between stochastic and systematic failure within ML models
- Probabilistic modeling of DNN outputs to mitigate performance and data deficits, such as inherent data uncertainty (aleatory uncertainty) or lack of data coverage (epistemic uncertainty)
- Introspective methods to analyze existing predictions
What are you particularly interested in?
- Technical reliability of AI systems
- Use of neural networks within critical applications, especially autonomous driving
- Uncertainty quantification for (deep) neural networks, and their use to mitigate existing uncertainties
- Performance analysis on input subsets, especially semantic in nature
Fouad Alkhoury
Research Focus:
Trustworthy ML
What problems are you currently working on?
-
- Efficient mining of frequent patterns
- Communication efficient distributed learning
What are you particularly interested in?
Development of interpretable methods that could increase the transparency of the black-box models in order to obtain more explainable solutions.
Dominik Baack
Research Focus:
Resource-aware ML
What problems are you currently working on?
I am currently working on accelerating the propagation of optical photons through the atmosphere in the context of CORSIKA 8 development. Photons consume, besides radio emissions, a large part of CPU time. By using specialized hardware accelerators like vectoring, GPUs or FPGAs, the runtime and energy consumption can be reduced significantly. It is important that the physical precision is maintained and the generated results remain reproducible.
What are you particularly interested in?
Use of modern hardware and methods to improve the overall efficiency of physical simulations
Alexander Becker
Research Focus:
Trustworthy ML and Forgetting in ML
What problems are you currently working on?
- Identifying and collecting existing approaches that can be considered as Forgetting in ML
- Transferring insights about cognitively different kinds of forgetting to the ML domain and elaborating their relevance for various fields of application
What are you particularly interested in?
Clarifying what it means to *forget* something on a cognitive level, axiomatizing general properties and practically applying these insights to state-of-the-art ML.
Katharina Beckh
Research Focus:
Human-oriented modeling, ML and Complex Knowledge
What problems are you currently working on?
- Modeling visual attention of anesthesiologists with focus on effort
- Steering active learning models with expert input
- Identification of bad medical reporting with text mining tools
What are you particularly interested in?
Interactive Machine Learning, improving the interaction between humans and machine learning algorithms by designing suitable interfaces and models.
David Biesner
Research Focus:
ML and Complex Knowledge
What problems are you currently working on?
- Interpretable NLP: Human readable approaches to word embeddings and topic modeling using matrix factorization
- Deep-learning based methods for domain-specific NLP, mainly text classification and text generation
What are you particularly interested in?
My main research focus is natural language processing (NLP). I am interested in the power of modern deep-learning architectures for general language understanding and transfer learning, and the possibilities of augmenting these approaches with human knowledge.
Ewald Bindereif
Research Focus:
Hybrid ML, Natural Language Processing
What problems are you currently working on?
Adapting text representations for causal inference using causal structures for more robust models
What are you particularly interested in?
The intersection between causal inference and Natural Language Processing, e.g., to identify and exploit causal effects in texts.
Daniel Boiar
Research Focus:
ML and Complex Knowledge
What problems are you currently working on?
- Building a knowledge base about multivariate time series algorithms
- Outlier detection with Support Vector Machines
What are you particularly interested in?
I am interested in derivation of new and reliable time series algorithms to optimize the analysis of heterogeneous sensor time series in an industrial context.
Eduardo Alfredo Brito Chacón
Research Focus:
Hybrid ML
What problems are you currently working on?
- Explainable semantic textual similarity functions for various information retrieval use cases e.g., retail product categorization, and fake news detection.
- Incorporating expert knowledge into NLP models, specifically to identify relevant sentences among legal documents.
What are you particularly interested in?
Developing informed Machine Learning models that can be to some extent explainable and resource-aware while staying competitive compared to their equivalent best performing deep learning models.
Kostadin Cvejovski
Awards:
Research Focus:
ML and Complex Knowledge
What problems are you currently working on?
- Improvement of recommender systems by using the text and time component of user-item interaction
- Use of deep neural networks to model queuing systems
What are you particularly interested in?
Development of methods that incorporate temporal and textual information to model how the inspection data of objects (products, companies, …) change over time.
Keerthana Dakshinamoorthy
Research Focus:
Applications of ML in Logistics
What problems are you currently working on?
- Application of ML techniques in quadcopters for Logistic industries
- Human activity recognition
Tobias Deußer
Research Focus:
ML and Complex Knowledge
What problems are you currently working on?
- Developing NLP models to understand the relations between text and accompanying tables and create meaningful numerical representations of table cells.
- Text generation leveraging the power of deep learning models
What are you particularly interested in?
Finding ways to fuse modern Natural Language Processing models and expert knowledge so that we can minimise the effective resource requirements of said NLP models.
Dr. Tiansi Dong
Awards:
Best Paper Award at ICANN 2019
Research Focus:
Trustworthy ML
What problems are you currently working on?
- Surveying recent papers on ‘Vision and Language’
- Reproduction of experiments in some of the papers
- Region-based knowledge graph reasoning
What are you particularly interested in?
I am interested in the unification of symbolic structures and in deep learning and its applications in Natural Language Understanding, knowledge graph reasoning and Visual-Language interaction.
Raphael Fischer
Research Focus:
Human-oriented modeling, ML and Complex Knowledge
What problems are yoou currently working on?
Application of probabilistic ML to spatio-temporal data, such as image time series, using generative ML to fill gaps in incomplete data sets, such as cloudy satellite image data
What are you particularly interested in?
To investigate how probabilistic ML can contribute to understanding our world while transferring knowledge and perception to other research areas and even to society in general.
Thore Gerlach
Research Focus:
Machine Learning on Quantum Computers
What problems are you currently working on?
- Quantum image processing for e.g. removal of clutter in the image
- Quantum kernel representations for the usage in classical kernel methods, e.g. Support Vector Machines (SVM) or Gaussian Processes (GP)
- Solving combinatorial optimization problems with adiabatic quantum computing
What are you particularly interested in?
Finding new ways to combine classical Machine Learning models with already known Quantum algorithms. The mathematical description as well as the testing of the performance on real Quantum Computers is of interest.
Sven Giesselbach
Research Focus:
ML and Complex Knowledge
What problems are you currently working on?
- Transferring knowledge from “labeled” domains to “unlabeled” domains with our algorithm named Corresponding Projections
- Including structured knowledge in text representations such as Word2Vec or BERT
What are you particularly interested in?
I think the assumption that computers can learn to understand natural language simply by consuming as many texts as possible is fundamentally flawed and I would like to explore mechanisms to incorporate external knowledge in natural language understanding.
Felix Gonsior
Research Focus:
Human-oriented ML, trustworthy ML
What problems are you currently working on?
- Interpreting optimization problems originating from training Machine Learning models as probabilistic models. This adds the possibility to evaluate and compare many possible alternative solutions.
- Inference via sampling on PGMs with quality guarantees for given sample sizes.
What are you particularly interested in?
Understanding the role of uncertainty in ML training and inference and it’s effect on the generalization ability of machine learning models.
Michael Gref
Research Focus:
Hybrid ML
What problems are you currently working on?
-
- Robust speech recognition for domains with little training data
- Multimodal emotion recognition
What are you particularly interested in?
Development of robust methods for automatic indexing and searching in audiovisual data bases – with practical application in industrial contexts or at the intersection with other research fields, such as humanities or history. In particular, I have a special interest in speech processing methods.
Vishwani Gupta
Lars Patrick Hillebrand
Research Focus:
ML and Complex Knowledge
What problems are you currently working on?
- Development of matrix factorization methods for joint learning of word embeddings and extraction of topics and their relations among each other.
- Metric learning for automatic evaluation of text summarization systems based on common human criteria.
What are you particulary interested in?
Combining structured knowledge and statistical learning in order to improve text representation learning and knowledge extraction (e.g. topics extraction, text summarization) in the context of Natural Language Processing and Understanding.
Matthias Jakobs
Research Focus:
Trustworthy ML
What problems are you currently working on?
- Providing guarantees in theory and application for explainability methods using Shapley values
- Combining explainability models with Bayesian Neural Networks (BNN)
What are you particularly interested in?
Shining a light into the decision-making process of black-box models giving users and experts confidence in the decisions produced by Neural Networks
Birgit Kirsch
Research Focus:
ML and Complex Knowledge
What problems are you currently working on?
- Cold-start knowledge base population by utilizing prior knowledge, e.g. ontologies, domain-specific constraints, …
- Noise reduction in distantly supervised datasets for relation extraction via statistical relational learning
What are you particularly interested in?
Joint learning of neural networks and logic based systems
Jens Leveling
Research Focus:
Resource-Constrained Learning
What problems are you currently working on?
- Resource-Constrained Learning for Logistics Use-Cases
- Industry Transfer of ML2R results – Main focus on Logistics Use-Cases
What are you particularly interested in?
Machine Learning on Internet-of-Things devices: Machine Learning for Classification, Detection and Tracking in Logistics Use-Cases
Industry transfer issues: Getting and creating training data shifting Machine Learning results on an operational level
Max Lübbering
Research Focus:
Hybrid Machine Learning
What problems are you currently working on?
I am trying to teach deep neural networks to know what they don’t know to make them deployable in safety-critical scenarios
What are you particularly interested in?
- Leveraging reconstructive representation learning for an increased model robustness
- Optimization / automation of Machine Learning workflows
Sascha Mücke
Research Focus:
Resource-Contrained ML, Quantum ML
What problems are you currently working on?
- Implementing and continually improving a FPGA-based hardware solver for Ising models
- Exploring applications of MAP solvers, specifically for analyzing probabilistic models
What are you particularly interested in?
Probabilistic graphical models; quantum-based optimization and its potential applications to machine learning
Sebastian Müller
Research Focus:
Trustworthy Machine Learning
What problems are you currently working on?
- A user-centered quantifying measures for quality assessment of explanations.
- Word sense disambiguation using a hybrid approach.
What are you particularly interested in?
Equipping ML Models with a combination of an abstract reasoning mechanism and complex knowledge to obtain an interpretable model that is able to produce situation dependent explanations.
Maximilian Otten
Research Focus:
Ressource-Aware ML
What problems are you currently working on?
- Applying ML-based Computer Vision to Industry & Research Projects
- Searching for industrial Use-Cases which can be solved by applying State-of-the-Art Machine Learning based computer vision algorithms.
- Allowing that these Algorithms can be used in resource constraint environments
What are you particularly interested in?
My focus lies on applying deep learning-based Object-Detection algorithms to various problems in logistics. Examples include the detection of objects in a warehouse to increase transparency of the ongoing processes and the use of Object-Detection as a trigger event for an emergency stop in case of a human-machine-conflict.
Andreas Pauly
Research Focus:
Resource-Constrained Learning, Theory of Modular Machine Learning
What problems are you currently working on?
- Transferring Machine Learning methods from research to industry
- Regularization of deep learning methods to reduce the required training resources
What are you particularly interested in?
I am interested in enhancing low-resource Machine Learning methods to enable an inexpensive industry transfer.
Maren Pielka
Research Focus:
Hybrid ML
What problems are you currently working on?
My main research focus is Natural Language Processing, specifically Natural Language Inference, Sentiment Analysis and Recommender Systems. At the moment I am working on building a system to find contradictions in German text.
What are you particularly interested in?
I am particularly interested in teaching algorithms to understand natural language and communication, especially challenging rhetorical stylistic devices such as irony and sarcasm as well as contradictions.
Christian Pionzewski
Research Focus:
Ressource-Aware ML, Trustworthy ML
What problems are you currently working on?
- Applying ML-based Computer Vision to Industry & Research Projects
- Creating modern and trustworthy software architectures for ML-based systems
What are you particularly interested in?
- Solving industrial challenges with ML-based computer vision
- Visualizing the inference process of ML models to understand what they learned and why they make certain decisions
Rajkumar Ramamurthy
Research Focus:
Hybrid Machine Learning
What problems are you currently working on?
- Casting structured prediction tasks in Natural Language Processing (NLP) space as Reinforcement Learning (RL) problems
- Human-in-the-loop learning of NLP tasks
What are you particularly interested in?
I am particularly interested in the active and reinforcement learning of NLP tasks through human interaction.
Till Hendrik Schulz
Research Focus:
ML and Complex Knowledge
What problems are you currently working on?
- Graph Kernels based on a distance measure using tree edit distances of local tree structures
- Computationally feasible graph pattern matching algorithms using constrained homomorphism
What are you particularly interested in?
Learning on graphs, particularly knowledge extraction from single networks or sets of graphs with provable bounds or guarantees
Max Schwarz
Awards:
Top placements in various international robotics competitions: RoboCup@Home, Amazon Robotics Challenges, DARPA Robotics Challenge, Mohammed Bin Zayed International Robotics Challenges
Research Focus:
Machine Learning with Complex Knowledge
What problems are you currently working on?
- Real-time perception of objects during robotic manipulation
- Learning object perception from very few examples
What are you particularly interested in?
Addressing the particular demands robotics places on Deep Learning Methods, such as few training examples, efficiency, and robustness
Florian Seiffarth
Research Focus:
ML and Complex Knowledge
What problems are you currently working on?
- Enriching abstract closure systems by additional information, such as distances, with the goal of efficient classification, e.g. achieving maximal separations in closure systems based on monotone linkage functions
- Neural network layers with weight sharing based on data dependent rules (expert knowledge), applications to graph and node classifications and learning on “complex” data structures
What are you particularly interested in?
Mathematical understanding of Machine Learning with applications to abstract closure systems, e.g. closed frequent item sets, concept lattices etc. and learning on graphs, e.g. graph neural networks based on data dependent rules.
Patrick Seifner
Research Focus:
Hybrid ML
What problems are you currently working on?
Approximating Markov jump processes using neural networks
What are you particularly interested in?
Modelling temporal processes, e.g. estimating past and future states of a process
Eike Stadtländer
Research Focus:
ML and Complex Knowledge
What problems are you currently working on?
- Learning closure systems, e.g. from transactional databases for application in frequent itemset mining
- Convexity conditions in graphs, e.g. for simultaneous clustering
What are you particularly interested in?
Bridging the gap between theory and practical algorithms to improve the interpretability of learning systems.
Hanxiao Tan
Research Focus:
Trustworthy Machine Learning
What problems are you currently working on?
Providing model-agnostic explainability methods on 3D point clouds data for better understandable visualisations of 3D neural networks
What are you particularly interested in?
More accurate, efficient and intuitive local/global explainability approaches for 3D deep neural networks
Vanessa Toborek
Research Focus:
Trustworthy ML, Hybrid ML
What problems are you currently working on?
- Structured learning algorithms for Natural Language Processing
- Use of prior knowledge for the explainability of ML algorithms
What are you particularly interested in?
The question of how to improve a Deep Learning model’s performance and explainability without solely relying on computational power and a posteriori explanations, respectively.
Dr.-Ing. Oliver Urbann
Awards:
Multiple awards in robot soccer, particularly world championship on RoboCup
Research Focus:
Resource-Contrained ML
What problems are you currently working on?
Machine Learning compiler
What are you particularly interested in?
Software engineering and Machine Learning in a hardware focus context, e.g. SIMD vectorization, caching, memory limits etc.
Laura von Rueden
Research Focus:
Machine Learning and Complex Knowledge, Trustworthy Machine Learning
What problems are you currently working on?
- Informed Machine Learning: A taxonomy and survey of integrating prior knowledge into learning systems
- Street-map based validation of semantic segmentation in autonomous driving
What problems are you particularly interested in?
The combination of data-based Machine Learning with knowledge-based modelling (hybrid AI)
Dorina Weichert
Research Focus:
Hybrid ML, ML with complex knowledge
What problems are you currently working on?
- Robust Bayesian Optimization
-
Application of MCTS for the optimization of production processes
What are you particularly interested in?
The application of ML-based optimization methods for real challenges.
Dr. Pascal Welke
Research Focus:
ML and Complex Knowledge
What problems are you currently working on?
- Efficient search of frequent subgraphs
- Core methods for graphs
- Analysis of ML models
Rahel Wilking
Research Focus:
Resource-Constrained Learning, Trustworthy Machine Learning
What problems are you currently working on?
- Identify common components of Machine Learning processes and how to implement them efficiently.
-
Division of model classes into components and options to recombine them.
What are you particularly interested in?
Make Machine Learning accessible to more people and to facilitate its responsible use.
Hazem Youssef
Research Focus:
Hybrid Machine Learning
What problems are you currently working on?
- Using multi-view 6D object pose estimation for localization of mobile robots
- Applying object tracking and Deep Learning in industrial and logistics settings
What are you particularly interested in?
I am particularly interested in the utilization of Machine Learning and computer vision in robotics applications.
Olga Zatsarynna
Awards:
Award of the Bonner Informatik Gesellschaft, 2021
Research Focus:
Machine Learning with Complex Knowledge
What problems are you currently working on?
- Anticipation of human actions in videos
-
Unsupervised learning of representations for video understanding
What are you particularly interested in?
I use deep learning methods to improve recognition and anticipation of human actions in the video data.