Forschungsbericht 2025

Data Science Foundations E-21

Leitung: Ay, Nihat
Oberingenieur: Eppe, Manfred
Gründungsdatum: 1.4.2021

Institut auf TORE
Institutswebsite

The Institute for Data Science Foundations has been founded in April 2021, as a central component of the growth strategy of the Hamburg University of Technology. It is part of the Hamburg Innovation Port (HIP), a technology and innovation site located at the port of Hamburg-Harburg.
The institutes’s research and teaching activities are concerned with fundamental methods for automated and data-based knowledge extraction. The institute pursues a holistic approach in which central aspects of intelligent systems are researched in a unified manner. In particular, concepts and methods from machine learning, learning in deep neural networks, reinforcement learning, and embodied intelligence are integrated. Mathematical theory development plays a central role here and is supported and guided by experimental work in a robotics lab. Of particular interest is the data-driven extraction of causal, as opposed to merely associational, knowledge. Decision making, action, and behaviour rely on this kind of knowledge and the coherent coupling of its various representational levels, ranging from the sub-symbolic to the symbolic level of representation. Methodologically, the institute has a strong focus on information theory and geometry, both integrated within the field of information geometry.

Publikationen

• Generalising thermodynamic efficiency of interactions: inferential, information-geometric and computational perspectives - Preprint
  Chen, Qianyang; Ay, Nihat; Prokopenko, Mikhail
  arXiv: 2509.10102 (2025)
  Publisher DOI
  
  
• Meta-world+: an improved, standardized, RL benchmark - Preprint
  McLean, Reginald; Chatzaroulas, Evangelos; McCutcheon, Luc; Röder, Frank; He, Zhanpeng; Yu, Tianhe; Julian, Ryan; Zentner, K. R.; Terry, Jordan; Woungang, Isaac; Farsad, Nariman; Castro, Pablo Samuel
  39th Conference on Neural Information Processing Systems, NeurIPS 2025
  Open Access | Publisher DOI
  
  
• Efficient optical coating design using an autoencoder-based neural network model - Journal Article
  Chattopadhyay, Utsa; Carstens, Florian; Steinecke, Morten; Kellermann, Tarik; Wienke, Andreas; Hartl, Ingmar; Ay, Nihat; Heyl, Christoph; Tünnermann, Henrik
  Journal of Physics: Photonics 8: 015007 (2025)
  Open Access | Publisher DOI
  
  
• Torsion of α-connections on the density manifold - Conference Paper
  Ay, Nihat; Schwachhöfer, Lorenz
  7th International Conference on Geometric Science of Information, GSI 2025
  Publisher DOI
  
  
• Wasserstein KL-divergence for Gaussian distributions - Conference Paper
  Datar, Adwait; Ay, Nihat
  7th International Conference on Geometric Science of Information, GSI 2025
  Publisher DOI
  
  
• Computational models of the emergence of self-exploration in 2-month-old infants - Conference Paper
  Spisak, Josua; Benad, Jan; Heidersberger, Johannes; Verschoor, Stephan; Lanillos, Pablo; Dongheui Lee; Eppe, Manfred; Wermter, Stefan; Hoffmann, Matej; Tcaci Popescu Sergiu
  IEEE International Conference on Development and Learning, ICDL 2025
  Publisher DOI
  
  
• Analyzing Multimodal Integration in the Variational Autoencoder from an Information-Theoretic Perspective - Conference Paper
  Langer, Carlotta; Georgie, Yasmin Kim; Ilja Porohovoj; Hafner, Verena Vanessa; Ay, Nihat
  IEEE International Conference on Development and Learning, ICDL 2025
  Publisher DOI
  
  
• On the natural gradient of the evidence lower bound - Journal Article
  Ay, Nihat; Oostrum, Jesse van; Datar, Adwait
  Journal of Machine Learning Research 26: 1-37 (2025)
  Open Access
  
  
• Dynamics-aligned latent imagination in contextual world models for zero-shot generalization - Preprint
  Röder, Frank; Benad, Jan; Eppe, Manfred; Banerjee, Pradeep Kr.
  39th Conference on Neural Information Processing Systems, NeurIPS 2025
  Open Access | Publisher DOI
  
  
• Final report of the project Learning Conversational Action Repair for Intelligent Robots (LeCAREbot) - Research Report
  Eppe, Manfred; Wermter, Stefan
  DFG form 3.06 – 01/23: 1-13 (2025)
  Open Access
  
  
• Convergence properties of natural gradient descent for minimizing KL divergence - Journal Article
  Datar, Adwait; Ay, Nihat
  Transactions on Machine Learning Research (7): 1-28 (2025)
  Open Access
  
  
• A concise mathematical description of active inference in discrete time - Review Article
  van Oostrum, Jesse; Langer, Carlotta; Ay, Nihat
  Journal of Mathematical Psychology 125: 102921 (2025)
  Open Access | Publisher DOI
  
  
• Wasserstein KL-divergence for Gaussian distributions - Preprint
  Datar, Adwait; Ay, Nihat
  arXiv:2503.24022 (2025)
  Open Access | Publisher DOI
  
  
• Scilab-RL: a software framework for efficient reinforcement learning and cognitive modeling research - Journal Article
  Benad, Jan; Röder, Frank; Eppe, Manfred
  SoftwareX 29: 102064 (2025)
  Open Access | Publisher DOI
  
  
• Systematic construction of continuous-time neural networks for linear dynamical systems - Journal Article
  Datar, Chinmay; Datar, Adwait; Dietrich, Felix; Schilders, Wil
  SIAM journal on scientific computing 47 (4): C820-C845 (2025)
  Publisher DOI
  
  
• Information geometry of the Otto metric - Journal Article
  Ay, Nihat
  Information Geometry 8: 209–232 (2025)
  Open Access | Publisher DOI
  
  
• Temporal shaping of UV picosecond pulses for photoinjectors - Conference Paper
  Ilia, Denis; Ay, Nihat; Cai, Meng; Chen, Ye; Große-Wortmann, Uwe; Hartl, Ingmar; Hillert, Wolfgang; Jiang, Yujiao; Klemps, Alexander; Mahnke, Christoph; Panuganti, Harsha; Pressacco, Federico; Tavakol, Hamed; Tünnermann, Henrik
  Conference on Lasers and Electro Optics Europe and European Quantum Electronics Conference, CLEO 2025
  Publisher DOI
  
  
• On the Natural Gradient of the Evidence Lower Bound - Journal Article
  Ay, Nihat; van Oostrum, Jesse; Datar, Adwait
  Journal of Machine Learning Research 26: 222 (2025)
  
  
• AI-driven design of high-performance optical thin film coatings for ultrafast lasers - Conference Paper
  Chattopadhyay, Utsa; Carstens, Florian; Wienke, Andreas; Hartl, Ingmar; Ay, Nihat; Heyl, Christoph; Tünnermann, Henrik
  Conference on Lasers and Electro Optics Europe and European Quantum Electronics Conference, CLEO 2025
  Publisher DOI
  
  
• Geometric learning of latent parameters with Helmholtz Machines - Doctoral Thesis
  Várady, Csongor-Huba
  Technische Universität Hamburg (2025)
  Open Access
  
  
• Unique Information Through the Lens of Channel Ordering: An Introduction and Review - Journal Article
  Banerjee, Pradeep Kumar
  Entropy 27 (1): 29 (2025)
  Open Access | Publisher DOI
  
  

Projekte

• KiMeKo - Aufbau einer norddeuschen KI-Med-Kollaborationsplattform
  Deutsches Zentrum für Luft- und Raumfahrt e. V. (DLR); Laufzeit 2024-2027
  Projektleitung: Ay, Nihat
  Co-Projektleitung: Schlaefer, Alexander
  Externe beteiligte Einrichtungen: UniTransferKlinik Lübeck GmbH; Universität zu Lübeck; Universitätsklinikum Hamburg-Eppendorf (UKE); University of Rostock; Universität Bremen; University Medicine Greifswald; VDI/VDE Innovation + Technik GmbH
  
  
• OPAL-FEL - Optimierte Laserpulse für Freie-Elektronen-Laser, Teilprojekt 4
  Bundesministerium für Forschung, Technologie und Raumfahrt (BMFTR); Laufzeit 2023-2026
  Projektleitung: Ay, Nihat
  Externe beteiligte Einrichtungen: Deutsches Elektronen-Synchrotron DESY; Amphos GmbH; Universität Hamburg
  
  
• DASHH Helmholtz Graduiertenkolleg
  Helmholtz-Zentrum Hereon; Laufzeit 2023-2026
  Projektleitung: Ay, Nihat
  Co-Projektleitung: Tünnermann, Hendrik
  
  
• SPP 2134: Informations-Integration in Prädiktiven Prozessen: Eine Mechanistische Verankerung des Selbst
  Deutsche Forschungsgemeinschaft (DFG); Laufzeit 2022-2025
  Projektleitung: Ay, Nihat
  Externe beteiligte Einrichtungen: Humboldt-Universität zu Berlin
  
  
• MoReSpace - Modeling a robot's peripersonal space and body schema for adaptive learning and imitation
  Deutsche Forschungsgemeinschaft (DFG); Laufzeit 2018-2026
  Projektleitung: Eppe, Manfred
  Externe beteiligte Einrichtungen: Universität Hamburg