Lamarr Lab Visits: 2025.1

18 Feb 2025, 10:00 → 20 Feb 2025, 16:30 Europe/Berlin

JvF25/2-232 - Meeting Room North (Lamarr/RC Trust Dortmund)

Welcome to the Lamarr Lab Visits! The Lab Visits provide a setting to network, exchange ideas and collaborate beyond the physical and disciplinary borders of the Lamarr locations and research areas.

In this edition of the event series, TU Dortmund University provides the meeting rooms. The program is filled with sessions and activities that have been proposed by anyone attending. Topics include specific research questions to be discussed, research area meetings, informal exchanges about novel scientific developments, and hard-skill training sessions. The program rounds off with a dinner program.

For days 1 and 3, we follow the established "free for all" format where the program is completely open and anyone had the opportunty to pitch/organize sessions. Day 2 is reserved for scientific and interdisciplinary areas meetings, organized by the scientific coordinators and chairs. 

Please refer to this event page for all information on the event. We look forward to seeing many of you very soon.

Program: Open for your contributions

Tuesday 18 February

Arrival

Location

Joint Kick-Off Session: Welcome & Opening

Location

13:00 Lunch Break (no cash, only card)

Location: https://maps.app.goo.gl/PRtUzCvZLtPdtSKt6

Architecture Search and Functional Composition: Scientific Session

Conveners: Dr Andrej Dudenhefner, Prof. Jakob Rehof

Learning and Explaining from Time Series Data: Scientific Session

Convener: Prof. Emmanuel Müller

1 Heitor Gomes: "Learning from Data Streams: Research and Practice"

In this talk, I will discuss the challenges and opportunities of applying machine learning to streaming data. To illustrate key concepts, I will introduce CapyMOA, a new open-source library designed for efficient real-time learning.

Speaker: Heitor Gomes (Victoria University of Wellington, New Zealand)

2 Probabilistic Approach: Anomaly and Concept Drift Detection in Data Streams

Speaker: Shubham Gupta

3 Amal Saadallah: Online Adaptive Local Interpretable Tree Ensembles for Time Series Forecasting

Thanks to their inherent interpretability, tree models are widely utilized in various learning tasks, including time series forecasting. However, single tree models often suffer from overfitting, limiting their applicability to real-world scenarios. To address this issue, ensembles of tree models are commonly employed. Yet, ensemble construction must account for the dynamic nature of time series, which can be subject to significant changes and the so-called concept drift phenomenon. In this paper, we develop local tree ensembles by specializing individual trees across specific regions in the input time series. We select the trees based on the most recent local pattern and manage their interdependence explicitly to ensure diversity in the ensemble. This is achieved through a carefully designed weighting schema. The trees are updated in an informed manner over time following concept drift detection in the time series. In addition, our approach supports explainability in forecasting tasks. Through extensive empirical evaluation on diverse real-world datasets, our method demonstrates comparable or superior performance to state-of-the-art approaches and several baseline methods.

Speaker: Amal Saadallah (Lamarr Institute-TU Dortmund)

Lab-visit-25-1.pptx

Physics-informed AI for Surrogate Models: Scientific Session

This meeting is supposed to bring together people interested in physics-informed AI, particularly for building surrogate models.
We are interested in discussing the following, non-complete, list of topics and applications:
- Making use of physics knowledge (e.g. differential equations) in building AI systems
- Neural surrogate models (e.g. neural operators, PINNs, etc.) for:
- Fast data generation and system exploration
- Control
- Event diagnosis (e.g. event detection, localization, classification, etc.)
- Inverse problems (e.g. inferring system's parameters from sensor measurements, etc.)
- Water distribution networks
- Fluid dynamics
- .....

Conveners: Dr André Artelt, Dr Michiel Straat, Thorben Markmann

PhysicsInformedML.pptx

SurrogateModeling.pdf

SymmetryInformedRL.pdf

WaterDistributionNetworks.pptx

[CANCELLED] Continual and Federated Learning Meetup: Scientific Session

Contributed Talks welcome

Convener: Prof. Hermann Blum

[CANCELLED] Explainable AI from a philosophical perspective: Scientific Session

Convener: Prof. Eva Schmidt

Hedy Lamarr Mentoring Program: Open Forum

Offering mentorship to young AI researchers is of utmost importance in advancing the Lamarr Institute and the overall field of AI research.

Hence in our session, we would like to enter into an open exchange with you with the aim of kickstarting the Lamarr Institute’s very first mentoring program. The latter targets up-and-coming researchers, with an additional focus on female and underrepresented groups within the Lamarr team.

We will be discussing needs and wishes pertaining to the program and will already collect subscriptions for interested participants and mentor volunteers.

Conveners: Ann-Kathrin Oster (Lamarr Institute at TU Dortmund University), Prof. Lucie Flek

Indico for Meetings and Events: Hard Skill Session

This session introduces Indico, the event management tool used at the Lamarr Institute, developed as an open-source project by CERN. Designed for meetings, workshops, and conferences, Indico streamlines event creation, participant registration, and scheduling. Key features will be demonstrated, and participants will be guided through the process of setting up and managing events efficiently. No prior experience with Indico is required.

Convener: Dr Jens Buß (Lamarr Institute, TU Dortmund University)

Introduction_to_Indico.pptx

4 Introduction to Indico

Speaker: Dr Jens Buß (Lamarr Institute, TU Dortmund University)

Slides

Lamarr Cluster for solving Science Tasks: Hard Skill Session

This meeting is used to discuss the current status of our compute cluster and how we want to use it in the future. We discuss aspects of the usage of the cluster, its intended use-case and discuss the roadmap of how we want to use the cluster.
If you have your own stories with our cluster and ideas on how to (better) utilize it, feel free to discuss with us :-)

Note: This meeting is intended to share/discuss ideas and not meant to be a comprehensive cluster tutorial, but rather to discuss
open questions. Please check out the official tutorial under https://gitlab.tu-dortmund.de/lamarr/lamarr-public/cluster if you are interested in getting started with the cluster

Current plan:
- Current statistics on the usage of the Dortmund cluster
- (Maybe) Current situation on the usage of the Bonn cluster?
- How to use the cluster: interactive vs batch jobs
- Future of the cluster:
- Possible use-cases for the cluster: Hyperparameter opt. with database hosting?
- Endpoint hosting for e.g. webpages? Use dedicated service nodes
- Storage becomes limited. How to deal with large amounts of data?
- Closing the usage-gap between Dortmund and Bonn Clusters

Conveners: Dominik Baack, Dr Sebastian Buschjäger (Lamarr Institute for ML and AI, TU Dortmund)

LinkedIn Essentials for Beginners: Hard Skill Session

Conveners: Katharina Peters (Lamarr Institute), Zoé Sanchez

Grant Proposals: Hard Skill Session

The session will provide an overview of the funding landscape in Germany. It will include a 30-45 minute presentation by Sabine Hunze (Research Support Service, TU Dortmund) discussing key aspects of the funding landscape and available funding opportunities. Following the presentation, there will be time for questions and a discussion where participants can share their experiences with previous application processes, including challenges faced and reasons for rejected applications. This session aims to facilitate knowledge sharing among applicants.

Register here

Conveners: Prof. Wolfgang Rhode (TU Dortmund), Zorah Lähner (University of Bonn)

5 Welcome

6 Talk: Exploring the Funding Landscape

Presentation by Sabine Hunze (Research Support Service, TU Dortmund) discussing key aspects of the funding landscape and available funding opportunities.

Speaker: Dr Sabine Hunze (TU Dortmund, Grant Office)

Overview_GermanFundingLandscape_Lamarr 02_2025 Hunze.pdf

7 Discussion + Q&A

Following the presentation, there will be time for questions and a discussion where participants can share their experiences with previous application processes, including challenges faced and reasons for rejected applications. This session aims to facilitate knowledge sharing among applicants.

Speaker: Sabine Hunze

Bocca Pizza Napoletana: Dinner (no cash, only card)

If you have registered for the evening event, you will find a colorful ticket in the back of your name tag. If it says Bocca, join the respective participant group for this restaurant.
Have a look at the restaurant here: https://maps.app.goo.gl/hV8udGSQoNDH1X6k7

If you are not content with the restaurant choice and would like to switch to the other option, find a person to switch tickets with.

Bocca at Google Maps

Pfefferkorn NY Steakhouse: Dinner

If you have registered for the evening event, you will find a colorful ticket in the back of your name tag. If it says Pfefferkorn, join the respective participant group for this restaurant.

Have a look at the restaurant here: https://maps.app.goo.gl/tuFVZSG95v8CQqKP9

If you are not content with the restaurant choice and would like to switch to the other option, find a person to switch tickets with.

Pfefferkorn at Google Maps

Wednesday 19 February

Arrival: Mingling

Embodied AI: Research Area Meeting

Planned Agenda:

Physical Embodied AI Monthly (10-11:30)
▶ Welcome Round
▶ Updates on Embodied AI
▶ Pitch Session: ICRA Publications
▶ Planning for Joint Publications

Break-Out Sessions: Micro Focus Groups (~11:45-13:00)
▶ Reinforcement Learning
▶ Physics Simulation
▶ Computer Vision

Convener: Julian Eßer

Hybrid Machine Learning Session: Research Area Meeting

This session consists of one discussion and 7 short talks by Hybrid-ML researchers

Convener: Ramsés Sanchéz

8 (Discussion) How can we better utilize Lamarr’s Hybrid-ML research area?

Researchers at Hybrid-ML tackle a wide range of applied and theoretical problems, characterized by different data modalities, such as time series, graphs, natural language, images, and their combinations. Solving these problems requires drawing from an equally broad spectrum of background knowledge, from abstract algebra and statistical physics to cognitive psychology. Yet, despite this diversity, many of our target problems share common underlying concepts and structures. At Hybrid-ML, we should recognize and leverage this common structure to our advantage.
This discussion session will focus on only one question: How can we do this — viz. To recognize and leverage these concepts and structures— effectively and precisely?

Speaker: Dr Ramsés Sanchéz (Lamarr institute, University of Bonn)

9 Diffusion models for Sketch Animation

Sketch-based Modeling and Animation are challenging problems due to the inherent ambiguity, style differences and lack of datasets. We explore how existing methods can be improved by integrating Diffusion Models for Video and 3D content generation. For this, Score-based Distillation Sampling, Optical Flow and alternative sketch representations are considered.

Speaker: Lio Schmitz

10 Laplace-Beltrami operator for 3D Gaussian splattings

With a growing interest in 3D Gaussian splatting, there comes a need for geometry processing applications directly on this new representation. In this work, we propose a formulation to compute the Laplace-Beltrami operator, a commonly used tool in geometry processing, directly on Gaussian splatting leveraging the Mahalanobis distance, and show its improvement in accuracy compared to point clouds and comparable accuracy to meshes.

Speaker: Hongyu Zhou

11 Towards Trustworthy Graph Anomaly Detection via Logical Constraints

Currently established graph anomaly detection methods predominantly operate in an unsupervised or self-supervised manner, assuming minimal anomaly contamination and relying solely on data-driven signals to infer notions of (a)normality. While these methods can, in theory, capture the relevant complex structural and attribute-based patterns, they typically do not allow for the meaningful incorporation of prior knowledge as guidance. This however is crucial, as alignment with existing prior knowledge is essential to ensure a suitable model of normality, that deviations are relevant, and to provide guarantees for known cases.

This talk presents our early work on integrating logical constraints into graph anomaly detection to address this limitation. Our approach begins by learning logic formulas that capture the normative behavior of nodes — effectively simulating constraints given by experts to guide the model. We further outline the next steps of our work, which will focus on incorporating these learned constraints into an established graph neural network framework for anomaly detection, guiding the detection process and providing guarantees for instances where the constraints apply.

Speaker: Tim Katzke

11:20 Break

20 minute break

12 Polyra swarm learning

My research focuses on ensemble methods for unsupervised learning tasks. Recently, I discovered a surprisingly effective approach for anomaly detection, which I named as a Polyra swarm. Upon further investigation, I found that Polyra exhibits a property analogous to the universal function approximation capability of neural networks. This insight has led me to explore an alternative paradigm where, instead of neural networks, we leverage Polyra swarms for learning tasks. Although this research is still in its early stages, and the talk will be very high level, such swarms offer a promising advantage in interpretability compared to traditional neural networks.

Speaker: Simon Klüttermann

13 MCBench: A Benchmark Suite for Monte Carlo Sampling Algorithms

In this talk I introduce MCBench, a benchmark suite designed to assess the quality of Monte Carlo (MC) samples. The benchmark suite enables quantitative comparisons of samples by applying different metrics, including basic statistical metrics as well as more complex measures, in particular the sliced Wasserstein distance and the maximum mean discrepancy. We apply these metrics to point clouds of both independent and identically distributed (IID) samples and correlated samples generated by MC techniques, such as Markov Chain Monte Carlo or Nested Sampling. Through repeated comparisons, we evaluate test statistics of the metrics, allowing to evaluate the quality of the MC sampling algorithms.

Speaker: Zeyu Ding

14 Concrete vs Abstract Planning for Large Language Models

Large language models are strong heuristic reasoners, but their planning abilities remain poor. We introduce a method for language models to learn to plan from unlabeled data by using a planner model to predict many steps ahead and conditioning the language model on the predicted plans. A crucial parameter in this framework is the level of abstraction of the generated plans: While some tasks arguably benefit more from high-level planning (e.g. creative writing), others require planning in the concrete language space (e.g. mathematical reasoning). In this talk, we explore both ends of the spectrum and finally ask the question if the right granularity can be learned from data.

Speaker: Florian Mai

15 Towards Beyond Black Box Distillation for VLLMs: Active Reasoning Transfer Using Synthetic Medical Data Generation

Speakers: Armin Berger, Helen Schneider

Resource-Aware ML: Research Area Meeting

This session will be split into two halves: In the first half, we present / discuss / review two contributions from Lamarr to resource-aware ML. In the second half, we will focus more on sustainability in computer science in general and how this might shape our research in resource-aware ML@Lamarr.

Convener: Sebastian Buschjäger (Lamarr Institute for ML and AI, TU Dortmund)

16 Splitting Stump Forests: Tree Ensemble Compression for Edge Devices

In this talk, we introduce Splitting Stump Forests – small ensembles of weak learners extracted from a trained random forest. The high memory consumption of random forest ensemble models renders them unfit for resource-constrained devices. We show empirically that we can significantly reduce the model size and inference time by selecting nodes that evenly split the arriving training data and applying a linear model on the resulting representation. Our extensive empirical evaluation indicates that Splitting Stump Forests outperform random forests and state-of-the- art compression methods on memory-limited embedded devices.

Note: This paper received the best paper award at Discovery Science 2024
Talk by Fouad Alkhoury

17 Rejection Ensembles with Online Calibration

As machine learning models become increasingly integrated into various applications, the need for resource-aware deployment strategies becomes paramount. One promising approach for optimizing resource consumption is rejection ensembles. Rejection ensembles combine a small model deployed to an edge device with a large model deployed in the cloud, with a rejector tasked to determine the most suitable model for a given input. Due to its novelty, existing research predominantly focuses on ad-hoc ensemble design, lacking a thorough understanding of rejector optimization and deployment strategies. In this talk, we focus on this research gap by presenting a theoretical investigation into rejection ensembles and proposing a novel algorithm for training and deploying rejectors based on these novel insights. We give precise conditions of when a good rejector can improve the ensemble's overall performance beyond the big model's performance, and when a bad rejector can make the ensemble worse than the small model. Second, we show that even the perfect rejector can overuse its budget for using the big model during deployment. Based on these insights, we propose to ignore any budget constraints during training but introduce additional safeguards during deployment. Experimental evaluation on 8 different datasets from various domains demonstrates the efficacy of our novel rejection ensemble outperforming existing approaches. Moreover, compared to standalone large model inference, we highlight the energy efficiency gains during deployment on a Nvidia Jetson AGX board.

Note: This work has been published and presented at the ECML-PKDD 2024.

Speaker: Sebastian Buschjäger (Lamarr Institute for ML and AI, TU Dortmund)

18 Optimum-preserving QUBO parameter compression

Quadratic unconstrained binary optimization (QUBO) problems are well-studied, not least because they can be approached using contemporary quantum annealing or classical hardware acceleration. However, due to limited precision and hardware noise, the effective set of feasible parameter values is severely restricted. As a result, otherwise solvable problems become harder or even intractable. In this talk, we study the implications of solving QUBO problems under limited precision. Specifically, we discuss that the problem’s dynamic range has a crucial impact on the problem’s robustness against distortions. We show this by formalizing the notion of preserving optima between QUBO instances and explore to which extend parameters can be modified without changing the set of minimizing solutions. Based on these insights, we introduce techniques to reduce the dynamic range of a given QUBO instance based on the theoretical bounds of the minimal energy value. An experimental evaluation on random QUBO instances as well as QUBO-encoded BinClustering and SubsetSum problems show that our theoretical findings manifest in practice. Results on quantum annealing hardware show that the performance can be improved drastically when following our methodology.

Presenter: Thore Gerlach

19 GENUSES - 100% Upcycling of Gaming Consoles

Can we give outdated gaming consoles a second life in research and teaching? With GENUSES, we upcycle every single component of old Playstation 4 consoles in order to let them serve as a cost-effective teaching kit.

There is a video, check it out: https://www.youtube.com/watch?v=9iUO86Y1t8w
Talk is done by Christian Hakert

20 Open Discussion: Resource-Aware ML vs. Sustainability in ML

A long, long time ago, in a far away land, some smart people thought about how to connect hardware and machine learning to make ML and hardware more resource-aware. At this time, the term "resource-aware ML" came along. While our roots go back some 10-15 years now, the term "resource-aware ML" only partially reflects the current trend in ML research. In fact, most new projects and ideas typically somehow incorporate the term "sustainability" or "green". And indeed, only a portion of our research directly connects to hardware, but there is also broader research concerning science communication, societal impact, management of resources (beyond a single machine) as well as ML workflow management.
In this open discussion round, I want to ask you:
What do you think about the term "resource-aware" ML? Is this still fitting? Do we need to "rebrand" ourselves? Do we want to move to other ideas?

Note: This is an open discussion. Input from everyone is welcome. No need to formally prepare anything. We are flexible with the time in our slot :-)

Speaker: Sebastian Buschjäger (Lamarr Institute for ML and AI, TU Dortmund)

Trustworthy AI & Human-Centered AI: Research Area Meeting

Conveners: Bahavathy Kathirgamanathan (Fraunhofer IAIS), Gennady Andrienko (Fraunhofer Institute IAIS), Jakob Rehof, Maram Akila (Lamarr / IAIS), Natalia Andrienko (Fraunhofer Institute IAIS)

21 Welcome

Speaker: Jakob Rehof

22 The Anatomy of Evidence - An Investigation into Explainable ICD Coding

Speaker: Katharina Beckh

23 Directional ExplainableA AI: The Problem of the Rabbit and the Duck

Speaker: Carina Newen

24 Does the model think as we expect?

Speakers: Gennady Andrienko (Fraunhofer Institute IAIS), Natalia Andrienko (Fraunhofer Institute IAIS)

RuleExplorer-short.pdf

25 Text as parameter interactive prompt optimisation for large language models

Speaker: H.S. Lin (HHU Düsseldorf)

11:35 Coffe Break

26 From Local to Global Explanations

Speaker: Maram Akila (Lamarr / IAIS)

27 Fast Linear Decomposition of ReLU Networks

Speaker: A. Baudzus

28 Reinforcement Learning from Self-feedback

Speaker: C. van Niekerk

29 Leveraging Human-Centered ML to create more Explainable ML models

Speaker: Bahavathy Kathirgamanathan (Fraunhofer IAIS)

30 Closing

13:00 Lunch Break (no cash, only card)

Location: https://maps.app.goo.gl/PRtUzCvZLtPdtSKt6

Industry and Production: Application Area Meeting

Convener: Felix Finkeldey

31 From data to design: challenges of data mining in manufacturing engineering

Speaker: Jim Bergmann

32 Reduction of experimental efforts for predicting milling stability affected by concept drift using transfer learning on multiple machine tools

Speaker: Rekha Prasad

33 Predicting machining stability with a quantum regression model

Speaker: Sascha Mücke (TU Dortmund)

34 Systematic evaluation schema for selecting optimal explainable AI method on nominal text data

Speaker: Rekha Prasad

Life Sciences: Application Area Meeting

Advancing Understanding: Explainable Artificial Intelligence at
Interfaces between Machine Learning, Life Sciences, and Philosophy.

Keynotes:

• Philosophical perspective of the concept of “explanation” and “hypothesis testing” in AI.
• Scientific hypothesis testing via molecular machine learning and exploration of causal
  relationships in explainable artificial intelligence.

Seminars:

- From objectual to explanatory understanding with AlphaFold2.

- Explaining molecular diffusion models.

Conveners: Elena Xerxa (University of Bonn), Prof. Jürgen Bajorath

boge-mosig.pdf

Natural Language Processing: Application Area Meeting

Convener: Akbar Karimi (University of Bonn)

NLP Group @ Lamarr Introduction.pdf

NLP @ Lamarr Feb 2025.pdf

35 Introduction & NLP/LLM collaboration pitches

36 Birds of a feather session: LLMs for scientific discovery: call for collaborations and topic synergies

Idea Pitches
Use cases of LLMs in scientific discovery - types of applications that LLMs and LLM agents can solve
LLMs in physics, biomedicine and beyond
Research questions and priorities

37 Roundtable discussion on social agents: Potential of LLMs for simulating social behavior

Introduction to LLMs for personal and social modeling
Lamarr Dagstuhl 2026 on Socially Intelligent AI Systems
(Lucie Flek, Tomer Ullman, Maarten Sap, Jenn Hu)
LLMs and the Theory of Mind

What are the research questions are you interested in that can be modeled with LLM social agents?
What are the benefits and risks of such modeling?
How should we prioritize the research questions?

38 LLM research in academia and industry: collaboration opportunities and challenges across our locations

What are the common goals / how to collaborate while keeping out KPIs?
How can we help the industry?
What do we need in Academia?
How can we share resources towards a common goal?

39 Open slot for other topics (future of LLMs, superalignment, AI safety…) / Wrap up

Looking into the future

Physics: Application Area Meeting

Convener: Dr Jens Buß (Lamarr Institute, TU Dortmund University)

40 Opening words

Speaker: Prof. Wolfgang Rhode (TU Dortmund)

41 Temporal Patterns in Sunspot Numbers: Analysis and Long-Term Prediction

The study of sunspot numbers is crucial for understanding solar activity and its impact on Earth's climate and space weather. This research analyzes the temporal patterns in historical sunspot data and develops predictive models for long-term forecasting. Using statistical and deep learning techniques, we identify key trends, periodicities, and anomalies in sunspot cycles. The proposed models are evaluated using error metrics such as RMSE and SMAPE to assess their predictive accuracy. Our findings provide insights into solar cycle variations and contribute to improving forecasts of solar activity for scientific and practical applications.

Speaker: Amal Saadallah (Lamarr Institute-TU Dortmund)

42 ErUM-Data Proposal: Anomaly Detection with Dataset Shift (ADDS)

Anomaly and signal detection is one of the most important use cases of machine learning (ML) both in scientific and in commercial applications. Anomalous signals are measured relative to an expected behavior of data, i.e., relative to the background or to the priors. Relevant examples of anomalies and signals in physics can be: an excess of gamma-rays near the center of our Galaxy (a possible signature of annihilating dark matter); an excess of unassociated gamma-ray sources for a particular range of source parameters (a new class of gamma-ray sources); and an excess of events in collider experiments for a particular value of invariant mass (a signature of a new particle, i.e., a particle beyond the Standard Model).

A confident detection of such anomalies or signals has a significant impact on scientific developments. The problem is that, in many cases, the priors against which the anomalies are evaluated have uncertainties. In ML, the anomaly is modeled as a difference between the distribution of the training dataset and the distribution of the target dataset, which is generally referred to as dataset shift. However, this shift can not only stem from the presence of an anomaly, but also from a change in the distribution of background events. Data analysis and ML methods must ensure that a plain change in the background distribution is not falsely mistaken for an actual anomaly or signal.

In many cases, distinguishing an anomalous signal from a change in the background requires domain knowledge to constrain the possible changes in the background, so that the remaining excess events can be attributed to a signal (or anomaly) in the data. Within ML, research questions concerning such an attribution for quantification methods are extensively studied. Quantification research is actively developed in the research communities of ML and computer science but, so far, has had little application in the ErUM fields.

The main goal of this project is to adapt existing ML methods and to further develop the methods of quantification learning for the detection and analysis of anomalies and signal in the presence of uncertain backgrounds for research questions from astroparticle physics. Subsequent possible uses comprise several ErUM fields and commercial applications.

Speaker: Mirko Bunse (Lamarr Institute, TU Dortmund University)

main.pdf

15:00 Coffee break

43 Geometry Optimization for Plasma Fusion Reactors

Plasma fusion has the potential to provide an efficient and safe energy source, however, several technological challenges still remain before fusion reactors can be realized at large scale. One promising direction is called stellarator in which the plasma is guided into a possibly complex equilibrium flow by magnetic fields. The optimal form for this flow is still unknown and only very sparse data about different configurations is known. The problem can be formulated as a geometry optimization problem of the surface of a torus. In this talk, I will give an introduction to the topic and present the technical challenges and open problems.

Speaker: Zorah Lähner (University of Bonn)

20250219-Lab Visits.pdf

44 ErUM-Data Proposal: Intelligent and Reliable Sensors for Scientific Exploration (SenSE)

In the scientific experiments nowadays, sensor setups and algorithms that process sensory data are usually statically configured and applicable only for specific scenarios. The SenSE project, as a joint effort of 4 PIs from CS and 4 PIs from Physics, addresses the research question: how can we use machine learning models to make sensors flexible and resilient with respect to changing environmental conditions in scientific experiments while still being efficient. The seminar will provide a short summary of the proposal and a status report of the internal Lamarr effort of sensor fusion.

Speaker: Jian-Jia Chen

Lamarr-SenSE25.pdf

45 ErUM-Data Proposal: HEIKI - Holistic event interpretation methods using AI techniques for particle physics

High-energy particle physics experiments are on the brink of facing significant challenges in reconstructing complex events due to increasing intensities and energies. The scientific aim of the presented work is to address the growing computational complexity of event reconstruction while enhancing efficiency and improving the precision of analyses in the ATLAS, LHCb, and Belle II experiments. To achieve this, the work has two primary objectives. First, it will advance machine learning (ML)-based algorithms for track and vertex reconstruction and event interpretation, integrating these methods into the reconstruction software of these experiments. Second, it will contribute to the development of cross-experiment platforms.

Speaker: Johannes Albrecht (TU Dortmund & LAMARR)

albrecht.pdf

46 ErUM-Data Proposal: Physics-LLM - Application of Large Language Models for Research Data Management in Physics

Within the ErUM communities, the amount, acquistion rate and diversity of data has rapidly increased over the last decade. While these challenges have been successfully addressed by individual communities, with respect to data analysis, the application of FAIR principles is lagging behind and the existing infrastructure does not allow for a swift and convenient publication of data. Smaller experiments are particularly affected and developments addressing the collection, reduction and analysis of data, as well as its storing, sharing and finding are urgently required to facilitate the transition from Big Data to Smart Data. The Physics-LLM consortium addresses the afore-mentioned challenges by utilizing LLMs in the RDM context. Particular focus is put on the development of an LLM-enhaced RDM toolkit and the use of open source models, e.g. Teuken 7B.

Speaker: Tim Ruhe (TU Dortmund)

Planning and Logistics: Application Area Meeting

Hey everyone,

in this session we are planning on identifying Big Challenges in planning and logistics. It's going to be an interactive format, focused on collaboration and discussion between natural scientists and domain experts. We would be very happy if you decide to join our session!

Agenda:
▶ Welcome and round of introduction​
▶ Impulse: What characterizes a Big Challenge?​ (Heike Horstmann & Moritz Roidl)

▶ Workshop: What are the Big Challenges behind your research topics?​
- Identification of Big Challenges: Brainstorming in smaller groups​
- Consolidation and clustering in big group​
- Ideation: Brainstorming in smaller groups​
- Why is this a big challenge? ​
- What are approaches to formalize this challenge?​
- Presentation and discussion​

▶ Discussion of organization​
- How do we organize collaborative groups to follow up on identified topics?​
- When should we schedule meetings and how often?​
▶ Conclusion

Convener: Anike Murrenhoff

2025-02-19_Session_Planning-and-Logistics_summary_upload.pdf

2025-02-19_What-constitutes-a-big-challenge 1.pdf

Reconstruction and Simulation of Digital Humans

In this short, interactive presentation we demonstrate our full-body avatar scanner, which is equipped with 60 cameras and reconstructs a digital twin (or virtual avatar) of a person in <10 minutes. In addition, we show how to manipulate the avatar’s body weight and how to estimate anatomical details (bones, muscles, fat).

For those interested (it is also a great way to spend the time between the previous program and the dinner reservation), simply gather outside of the Lamarr building. Prof. Mario Botsch will walk you to the Lab.

Convener: Prof. Mario Botsch

Maximilian: Dinner

If you have registered for the evening event, you will find a colorful ticket in the back of your name tag. If it says Maximilian, join the respective participant group for this restaurant.

Have a look at the restaurant here: https://maps.app.goo.gl/kvogU4yvsq9YLqfA8

If you are not content with the restaurant choice and would like to switch to the other option, find a person to switch tickets with.

Maximilian at Google Maps

Wenkers am Markt: Dinner

If you have registered for the evening event, you will find a colorful ticket in the back of your name tag. If it says Wenkers am Markt, join the respective participant group for this restaurant.

Have a look at the restaurant here: https://maps.app.goo.gl/LPrjuKgP1tzAaa5z6

If you are not content with the restaurant choice and would like to switch to the other option, find a person to switch tickets with.

Wenkers at Google Maps

Thursday 20 February

Arrival: Mingling and Handling Luggage

Meet the RC Trust: Scientific Session

We would like to introduce our research center to you. Hence, if you want to meet and network with us beforehand, we start at 10:00 in the same lecture hall (E23 in OH14) with four introductory talks by Prof. Daniel Neider, Prof. Alexander Marx, Jan Corazza and Prof. Emmanuel Müller.

Agenda:

• Emmanuel Müller: Vorstellung RC Trust und dann Forschung zum Thema "Trustworthy Machine Learning"

• Daniel Neider: "Neural Networks with Safety Nets: A Neuro-Symbolic Framework for Neural Network Verification"

• Alexander Marx: How Causality can Contribute to Trustworthy AI

• Jan Corazza: "Formal Methods in Reinforcement Learning"

Convener: Emmanuel Müller

47 AI Colloquium: The road to AI-based discovery

Lecture in the AI Colloquium by Prof. Gregor Kasieczka (Universität Hamburg)

Abstract: Machine learning and AI have quickly turned into indispensable tools for modern particle physics. They both greatly amplify the power of existing techniques - such as supercharging supervised classification - and enable qualitatively new ways of extracting information - such as anomaly detection for unsupervised discovery. After briefly introducing the environment of collider based particle physics, this talk will review key developments and new directions in machine learning applied to data analysis.

Speaker: Prof. Gregor Kasieczka (Universität Hamburg)

AI Colloquium Link (Indico)

DeepLearning_Lamarr_2025.pdf

Lecture Recording

AI for Mathematics and Mathematics for AI / Foundation Inference Models for SDEs: Scientific Session

This joint session consists of the following blocks:

• AI for Mathematics and Mathematics for AI
  Speakers: Benno Kuckuck, Julian Kranz, Adrian Riekert, Arnulf Jentzen from the University of Münster.
  
  

• Discussion.
  
  

• Foundation Inference Models for SDEs: Towards AI Agents that Build and Reason about Mathematical Models of Data.
  Speaker: Ramses J. Sanchez
  Abstract: AI agents augmented with LLMs are becoming commonplace. In the near future, LLM-based agents for mathematical modeling will require access to fast and accurate tools, to infer mathematical equations they can reason about directly from data. However, inferring equations from data — also known as the inverse problem or system identification problem — remains a fundamental and open challenge in machine learning. Current state-of-the-art solutions are slow, unstable, and heavily reliant on prior knowledge, making them inaccessible to scientists unfamiliar with ML — let alone AI agents(!). In this presentation, I will discuss Foundation Inference Models, a methodology we have developed over the past two years that enables zero-shot inference of equations directly from data. This time, I will focus on Stochastic Differential Equations (SDEs), which serve as flexible mathematical frameworks for modeling complex systems across disciplines, from statistical physics to finance and climate science.
  
  

• Discussion

Conveners: Dr Adrian Riekert, Prof. Arnulf Jentzen, Dr Benno Kuckuck, Dr Julian Kranz, Dr Ramsés Sanchéz (Lamarr institute, University of Bonn)

Education in Lamarr: Current situation, new ideas and future outlook

The meeting aims to discuss various aspects of the education area within the Lamarr Institute.

Convener: Vanessa Faber (TU Dortmund)

WestAI: Boosting ML Trainings with HPC Resources

Abstract:
The capabilities of Machine Learning (ML) models usually scale with model size. However, this expansion also entails a need for greater computational resources. The AI service center WestAI addresses this by providing high performance computing (HPC) hardware for large model trainings in academia and industry. WestAI’s primary offerings include 10,000 hours on NVIDIA H100 GPUs and specialized ML consulting.

This session will cover:
- An overview of WestAI and how its services can support your work.
- Instruction on how to apply for computing time.
- An in-depth look at the HPC systems at RWTH Aachen University and the Jülich Supercomputing Centre.
- How to utilize the HPC system at RWTH with Jupyter Notebooks for small-scale testing and training.
- Instructions on how to use the batch system for large model trainings at RWTH’s HPC system.

The session will include ample time for questions and discussions.

Convener: John Arnold

WestAI_Lamarr_Lab_Visit.pdf

13:00 Lunch Break (no cash, only card)

Location: https://maps.app.goo.gl/PRtUzCvZLtPdtSKt6

UA Ruhr Distinguished Lecture Series: Presentation by Edward Lee

https://cs.tu-dortmund.de/en/details/ua-ruhr-distinguished-lecture-series-vertrauenswuerdige-ki-49444/

Location

Zoom Link (If you did not register for a seat)

48 Welcome by the President of TU Dortmund University

Speaker: Prof. Manfred Bayer (President of TU Dortmund University)

49 Introduction

Speaker: Emmanuel Müller