PeopleofSuperintelligence

Researcher

Jonas Köhler

Also known as Jonas Koehler

Jonas Köhler is a machine learning researcher whose work spans equivariant networks, normalizing flows and molecular simulation. While studying for a master's degree at the University of Amsterdam, he co-authored Spherical CNNs with Taco Cohen, Mario Geiger and Max Welling, which won the Best Paper Award at ICLR 2018. In 2019, Science published Boltzmann Generators, a method he co-authored with Frank Noé, Simon Olsson and Hao Wu as a PhD student at Freie Universität Berlin, and he went on to publish work on equivariant, stochastic and smooth normalizing flows. He was a Senior Researcher at Microsoft Research from 2023 to 2024. Since September 2024, he is a member of technical staff at CuspAI, the developer of the open-source kUPS simulation engine.

Now at
CuspAI
Badges

Timeline

  1. 2014–2018

    Studies in Weimar and Amsterdam

    He studies applied computer science at Bauhaus-University Weimar from 2014 to 2016 for a B.Sc., then artificial intelligence at the University of Amsterdam from 2016 to 2018 for an M.Sc. In 2018, he is also a research intern at MPI-IS Tübingen.

  2. 2018

    A best paper at ICLR

    With Taco Cohen, Mario Geiger and Max Welling, he co-authors Spherical CNNs, which introduces building blocks for convolutional networks that analyze spherical images, with uses such as omnidirectional vision for drones, robots and autonomous cars. The paper wins the Best Paper Award at ICLR 2018.

  3. 2019

    Boltzmann Generators in Science

    In 2019, Science publishes Boltzmann Generators, which he co-authors with Frank Noé, Simon Olsson and Hao Wu at Freie Universität Berlin, where he has been a PhD student in applied mathematics since December 2018. The method uses neural networks to turn a complex equilibrium distribution into one that is easy to sample, and the university presents it as a fundamentally new solution to the sampling problem in statistical physics.

  4. 2020

    Equivariant and stochastic flows

    In 2020, he co-authors Equivariant Flows with Leon Klein and Frank Noé for ICML; it shows that equivariant normalizing flows generate distributions that are invariant to the symmetries built into them. At NeurIPS 2020, Stochastic Normalizing Flows, written with Hao Wu and Noé, is a spotlight paper.

  5. 2021

    Smooth Normalizing Flows at NeurIPS

    In 2021, he co-authors Smooth Normalizing Flows with Andreas Krämer and Frank Noé for NeurIPS; the flows allow training by force matching to simulation data and serve as potentials in molecular dynamics simulations. From September to December 2021, he is a research scientist intern at DeepMind.

  6. January 2023–September 2024

    Senior Researcher at Microsoft Research

    He joins Microsoft Research as a Senior Researcher in January 2023 and stays until September 2024. In 2023, he co-authors flow-matching with Yaoyi Chen, Andreas Krämer, Cecilia Clementi and Frank Noé, a training method for coarse-grained molecular force fields that uses normalizing flows and captures the folding and unfolding transitions of small proteins. It appears in the Journal of Chemical Theory and Computation.

  7. Since September 2024

    Member of technical staff at CuspAI

    In September 2024, he joins CuspAI as a member of technical staff. In June 2026, he is a speaker at the dida conference in Berlin, with a talk on kUPS, an open-source JAX-native simulation engine developed at CuspAI in which Monte Carlo, molecular dynamics and relaxation compose freely with classical or learned potentials.

Career

  1. Since 2024CuspAIMember of technical staff
  2. 2023–2024Microsoft ResearchSenior Researcher
  3. 2021Google DeepMindResearch scientist intern
  4. 2018Max Planck Institute for Intelligent SystemsResearch intern

Education

  1. 2018University of AmsterdamM.Sc.
  2. From 2018Freie Universität BerlinPhD
  3. 2016Bauhaus-Universität WeimarB.Sc.

From the biography.

  1. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Jonas Köhler. The median h-index is 48 at universities and 14 at frontier AI labs. We compare publishing careers, recent citations, dual appointments and job moves across 2,420 university researchers and 308 frontier-lab researchers.
  2. 16% of AI career moves in the 2020s went from a university to a companyIncludes Jonas Köhler. We estimate that 16% of AI career moves in the 2020s went from a university to a company; 29% of moves starting at a university went to a company, and 76% ended the university role.
  3. Among employers with 30+ full-time cases, OpenAI leads primary technical role H-1B pay at $310,000, and Amazon, Google and Microsoft file 57% of all casesIncludes Microsoft. Across 440,669 certified US H-1B filings from 44 employers (FY2020–FY2026), OpenAI leads primary technical role base pay at $310,000 and xAI leads overall base pay at $290,000.
  4. Google had the most people shaping AI every year from 2014 to 2025Includes Google. Our estimates of which organizations employed the most people shaping AI each year from 1955 to 2025, tracing the shift from university labs to companies.

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 Jonas Köhler yet. Send us a free-licensed photo.

  1. Jonas Köhler argmin.xyz. Checked 21 October 2026.
  2. [1801.10130] Spherical CNNs arxiv.org. Checked 21 October 2026.
  3. 10 Cutting Edge Research Papers In Computer Vision & Image Generation topbots.com. Checked 21 October 2026.
  4. [1812.01729] Boltzmann Generators -- Sampling Equilibrium States of Many-Body Systems with Deep Learning arxiv.org. Checked 21 October 2026.
  5. Artificial Intelligence for Physics Research • Information for Media and Journalists • Freie Universität Berlin fu-berlin.de. Checked 21 October 2026.
  6. Equivariant Flows: Exact Likelihood Generative Learning for Symmetric Densities proceedings.mlr.press. Checked 21 October 2026.
  7. Smooth Normalizing Flows proceedings.neurips.cc. Checked 21 October 2026.
  8. [2203.11167] Flow-matching -- efficient coarse-graining of molecular dynamics without forces arxiv.org. Checked 21 October 2026.
  9. dida conference 2026 dida.do. Checked 21 October 2026.