PeopleofSuperintelligence

Researcher

Jan Tauberschmidt

Jan Tauberschmidt is a PhD student in the Data Science and its Applications department at DFKI, where he works on generative modeling, physics- and domain-informed learning, and control for scientific problems. In August 2025, he posted a preprint on fine-tuning flow-matching models to enforce physical constraints, and the paper was published at ICLR 2026 in Rio de Janeiro. He co-authored a reinforcement-learning preprint on race-car control with Jan Peters of TU Darmstadt and others in April 2026. In September 2026, he led a preprint on continuous-time generative dynamics on learned data manifolds, and in October 2026 he posted another on learning PDE dynamics with Green's Observation Operators.

Timeline

  1. August 2025

    Physics-constrained flow matching

    In August 2025, he posts the preprint "Physics-Constrained Fine-Tuning of Flow-Matching Models for Generation and Inverse Problems" with Sophie Fellenz, Sebastian J. Vollmer and Andrew B. Duncan. The framework fine-tunes flow-matching generative models to enforce physical constraints and solve inverse problems in scientific systems.

  2. April 2026

    A paper at ICLR 2026

    The flow-matching paper is published at ICLR 2026, the Fourteenth International Conference on Learning Representations, held in Rio de Janeiro in April 2026.

  3. April 2026

    Race-car control with reinforcement learning

    In April 2026, he is second author of the preprint "Behavior-Constrained Reinforcement Learning with Receding-Horizon Credit Assignment for High-Performance Control", written with Siwei Ju, Oleg Arenz, Peter van Vliet and Jan Peters. Peters, Ju and Arenz work at the Intelligent Autonomous Systems Lab of TU Darmstadt, and the learned policies are tested in high-fidelity race car simulation with data from professional drivers.

  4. September 2026

    Walking the score manifold

    In September 2026, he is first author of "Walking the Score Manifold: Continuous-time Generative Dynamics on Learned Data Manifolds", written with Brian B. Moser, Stanislav Frolov, Andreas Dengel, Andrew B. Duncan and Sebastian J. Vollmer. The method uses pretrained score-based models as geometric priors and learns a vector field that evolves data along score-induced interpolation paths.

  5. October 2026

    Green's Observation Operators

    In October 2026, he posts "Learning PDE Dynamics between Submanifolds Using Green's Observation Operators" with seven co-authors, among them Sophie Fellenz, Marius Kloft, Jonas Latz and Sebastian Josef Vollmer. The Green's Observation Operator maps the ambient medium once to the Green's kernel of a linear PDE restricted to the source and observation submanifolds, so new sources cost one lower-dimensional integral and no network evaluation.

  6. As of October 2026

    PhD student at DFKI

    He is a PhD student in the Data Science and its Applications department at DFKI. His research covers generative modeling, physics- and domain-informed learning, and control for scientific problems.

  1. Nonprofit roles are 2.3 times as common in AI safetyIncludes DFKI. We estimate where people focused on safety, alignment and evaluation work. Nonprofit roles appear for 25.1% of safety people with known affiliations, compared with 11.0% in other areas.
  2. The top 10 organizations employed 22% of people shaping AI in 2025: 51% of company researchers and 21% of university researchersIncludes DFKI. We estimate the share of people shaping AI who worked at the top 10 organizations each year from 1955 to 2025, comparing concentration across companies, universities, and institutes and labs.
  3. The United States employs an estimated 48% of people shaping AIIncludes DFKI. We estimate where the people shaping AI work today, how universities, companies and research institutes split across countries, and how country shares have changed since the 1960s.
  4. More people shaping AI studied at Massachusetts Institute of Technology than at any other institutionIncludes DFKI. We estimate where the people shaping AI studied and where their careers led. Massachusetts Institute of Technology leads with 467 former students, including 291 recorded doctorates.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Jan Tauberschmidt yet. Send us a free-licensed photo.

  1. [2508.09156] Physics-Constrained Fine-Tuning of Flow-Matching Models for Generation and Inverse Problems arxiv.org. Checked 21 October 2026.
  2. Jan Tauberschmidt, M.Sc. www-live.dfki.de. Checked 21 October 2026.
  3. Tauberschmidt, Jan | ML Anthology mlanthology.org. Checked 21 October 2026.
  4. [2604.03023] Behavior-Constrained Reinforcement Learning with Receding-Horizon Credit Assignment for High-Performance Control arxiv.org. Checked 21 October 2026.
  5. Behavior-Constrained Reinforcement Learning with Receding-Horizon Credit Assignment for High-Performance Control (HTML version) arxiv.org. Checked 21 October 2026.
  6. [2609.17901] Walking the Score Manifold: Continuous-time Generative Dynamics on Learned Data Manifolds arxiv.org. Checked 21 October 2026.
  7. [2610.01697] Learning PDE Dynamics between Submanifolds Using Green's Observation Operators arxiv.org. Checked 21 October 2026.
  8. Jan Tauberschmidt — Data Science and its Applications dsa.dfki.de. Checked 21 October 2026.