We study the principles that make machine learning work — and develop methods that remain reliable when data are limited, distributed, biased, or high-stakes. Our research spans deep learning theory, causal machine learning, trustworthy AI, and privacy-preserving learning.
Kamp Lab-Machine Learning and Artificial Intelligence Group at TU Dortmund University
The Kamp Lab conducts research on trustworthy and reliable machine learning, ranging from methodological foundations to applications in sensitive and high-stakes domains. Our work focuses on four main areas: (i) federated and privacy-preserving learning, (ii) deep-learning theory and generalization, (iii) causal machine learning, and (iv) trustworthy AI in medicine and other high-stakes domains. We develop new machine-learning methods, study their theoretical and empirical properties, and investigate how they can be applied responsibly to real-world problems.
Research Areas
Kamp Lab
Machine Learning and Artificial Intelligence at TU Dortmund University
Kamp Lab is the Machine Learning and Artificial Intelligence research group at TU Dortmund University, led by Prof. Dr. Michael Kamp. We develop principled machine-learning methods for settings in which data are distributed, sensitive, heterogeneous, or difficult to centralize and where automated decisions require particular care.
Our work combines fundamental machine-learning research with interdisciplinary applications. We study the foundations of learning and generalization as well as federated, causal, and trustworthy machine learning, with particular interest in medical and other high-stakes environments.
Our Philosophy
Understand, not just predict
We want to understand why machine-learning methods work, when they fail, and what makes their predictions reliable. This means going beyond benchmark performance toward mechanisms, mathematical explanations, causal structure, and theoretical guarantees whenever possible.
Fundamental research with a purpose
Not every result needs an immediate application, but we choose research directions with real-world relevance in mind. Healthcare is particularly important to us: it is a domain where AI can genuinely help people and where questions of reliability, robustness, causality, privacy, and generalization become unavoidable.
Independence and ambition
We want to develop excellent, independent researchers. We give people substantial freedom and support their scientific and professional development, while expecting intellectual ownership, initiative, and high standards in return.
Research Environment
Our group is based at TU Dortmund University and closely embedded in the Lamarr Institute for Machine Learning and Artificial Intelligence (https://lamarr-institute.org/). At the same time, we are closely integrated with the Institute for Artificial Intelligence in Medicine (IKIM) at University Medicine Essen (https://www.ikim.uk-essen.de/). Michael Kamp is a researcher at IKIM, where he established the Trustworthy Machine Learning research group, and our doctoral researchers are affiliated with IKIM as guest researchers..
Our research spans two complementary environments: foundational machine-learning research at TU Dortmund and the Lamarr Institute, and medical AI at the Institute for Artificial Intelligence in Medicine (IKIM) at University Medicine Essen.
Recent Highlights
- I am super proud of my group getting one conference paper and 6 workshop papers accepted at NeurIPS 2026. Great work by Amr Abourayya, Rania Briq, Ting Han, Osman Mian, and Andy Tai.
- My PhD student Amr Abourayya, my colleague Jens Kleesiek, and I published a paper on “Beyond Parameter Aggregation: Semantic Consensus for Federated Fine-Tuning of LLMs” at NeurIPS 2026 (A*, top 7%)
- My PhD student Rania Briq, my colleagues Ohad Fried, Sarel Cohen, Stefan Kesselheim, and I published a paper on “Exploring and Exploiting Stability in Latent Flow Matching” at ICML 2026 (A*, top 7%)
- I am terribly proud of my postdoc Osman Mian for receiving the AAAI Outstanding Paper Award 2026 for his work on “Causal Structure Learning for Dynamical Systems with Theoretical Score Analysis“.
- My postdoc Osman Mian, my colleague Jens Kleesiek, and I published a paper on “Unified Causal Discovery and Missing Data Imputation” at AISTATS 2026 (A, top 13%).
- Our project FLIP-IT for federated learning in a network of general practitioners started in January, 2026. The project is funded by KI.NRW (EFRE) with the goal to build a trustworthy and secure federated learning infrastructure in GP practices. With this, we will train early warning models for chronic kidney disease to protect patients from kidney failure and dialysis.
- My PhD student Ting Han, my colleagues Linara Adilova, Henning Petzka, Jens Kleesiek, and I published a paper on “Flatness is Necessary, Neural Collapse is Not: Rethinking Generalization via Grokking” at NeurIPS 2025 (A*, top 7%)
Explore the Kamp Lab
Learn more about our popular repositories, and ongoing research activities.
News
Little is Enough: Boosting Privacy in Federated Learning with Hard Labels
Layer-Wise Linear Mode Connectivity
Federated Daisy-Chaining
Nothing but Regrets – Federated Causal Discovery
